International Journal of Computational Science and Engineering (69 papers in press)

Regular Issues

• BITSAT: an efficient approach to modern image cryptography  
  by Sheel Sanghvi, Ramchandra Mangrulkar 
  Abstract: This paper proposes a new approach towards efficient image cryptography that works using the concepts of bit-plane decomposition and bit-level scrambling. This method does not require the involvement of additional or extra images. Users have the flexibility of choosing (1) any bit-plane decomposition method; (2) logic that runs the key generation block; (3) customisation in the bit-level permutations performed during scrambling. The implementation of the method is simple and free of heavily complex steps and operations. This makes the algorithm applicable in a real-world scenario. The BITSAT algorithm is applied to a variety of images and, consequently, the encrypted images generated showcase a high level of encryption. The analysis and evaluation of the algorithm and its security aspects are performed and described in detail. The paper also presents the application domains of the method. Overall, the results and analysis indicate a positive working scope and suitability for real-life applications.
  Keywords: hybrid image encryption; image cryptography; novel image encryption scheme; bit-level scrambling; bit-plane decomposition.
  DOI: 10.1504/IJCSE.2022.10049691
   
  
• Multiple correlation based decision tree model for classification of software requirements  
  by Pratvina Talele, Rashmi Phalnikar 
  Abstract: Recent research in Requirements Engineering (RE) includes requirements classification, and use of Machine Learning (ML) algorithms to solve RE problems. The limitation of existing techniques is that they consider only one feature at a time to map the requirements without considering the correlation of two features and are biased. To understand these limitations, our study compares and extends the ML algorithms to classify requirements in terms of precision and accuracy. Our literature survey shows that decision tree (DT) algorithm can identify different requirements and outperforms existing ML algorithms. As the number of features increases, the accuracy using the DT is improved by 1.65%. To overcome the limitations of DT, we propose a Multiple Correlation Coefficient based DT algorithm. When compared with existing ML approaches, the results showed that the proposed algorithm can improve classification performance. The accuracy of the proposed algorithm is improved by 5.49% compared with the DT algorithm.
  Keywords: machine learning; requirements engineering; decision tree; multiple correlation coefficient.
  
  
• Improved performance on tomato pest classification via transfer learning based deep convolutional neural network with regularisation techniques  
  by Gayatri Pattnaik, Vimal K. Shrivastava, K. Parvathi 
  Abstract: Insect pests are major threat to the quality and quantity of crop yield. Hence, early detection of pests using a fast, reliable and non-chemical method is essential to control the infestations. Hence, we have focused on tomato pest classification using pre-trained deep convolutional neural network (CNN) in this paper. Four models (VGG16, DenseNet121, DenseNet169, and Xception) were explored with transfer learning approach. In addition, we have adopted two regularization techniques viz. early stopping and data augmentation to prevent the model from overfitting and improve its generalization ability. Among four models, the DenseNet169 achieved highest classification accuracy of 95.23%. The promising result shows that the DenseNet169 model with transfer learning and regularization techniques can be used in agricultural for pest management.
  Keywords: agriculture; convolutional neural network; data augmentation; early stopping; pest; regularisation; tomato plants.
  DOI: 10.1504/IJCSE.2023.10054281
   
  
• A collaborative filtering recommendation algorithm based on DeepWalk and self-attention  
  by Jiaming Guo, Hong Wen, Weihong Huang, Ce Yang 
  Abstract: Graph embedding is one of the vital technologies in solving the problem of information overload in recommendation systems. It can simplify the vector representations of items and accelerates the calculation process. Unfortunately, the recommendation system using graph embedding technology does not consider the deep relationships between items when it learns embedding vectors. In order to solve this problem, we propose a collaborative filtering recommendation algorithm based on DeepWalk and self-attention in this paper. This algorithm can enhance the accuracy in measuring the similarity between items and obtain more accurate embedding vectors. Chronological order and mutual information are used to construct a weighted directed relationship graph. Self-attention and DeepWalk are used to generate embedding vectors. Then item-based collaborative filtering is used to obtain recommended lists. The results of the relative experiments and evaluations on three public datasets show that our algorithm is better than the existing ones.
  Keywords: DeepWalk; self-attention; mutual information; collaborative filtering; recommendation algorithm.
  DOI: 10.1504/IJCSE.2022.10050515
   
  
• An aeronautic X-ray image security inspection network for rotation and occlusion  
  by Bingshan Su, Shiyong An, Xuezhuan Zhao, Jiguang Chen, Xiaoyu Li, Yuantao He 
  Abstract: Aviation security inspection needs lots of time and human labour. In this paper, we establish a new network for detecting prohibited objects in aeronautic security inspection X-ray images. Objects in the X-ray image often present rotated shapes and overlap heavily with each other. In order to solve the rotation and occlusion in X-ray image detection, we construct the De-rotation-and-occlusion Module (DROM), an efficient module that can be embedded into most deep learning detectors. Our DROM leverages the edge, colour and Oriented Fast and Rotated BRIEF (ORB) features to generate an integrated feature map, while the ORB features could be extracted quickly and diminish the deviation produced by rotation effectively. Finally, we evaluate DROM on the OPIXRay dataset; compared with several latest approaches, the experimental results certify that our module promotes the performance of Single Shot MultiBox Detector (SSD) and obtains higher accuracy, which proves the modules application value in practical security inspection.
  Keywords: X-ray image detection; de-rotation-and-occlusion; deep learning; ORB; SSD.
  
  
• ReFIGG: retinal fundus image generation using generative adversarial networks  
  by Sharika Sasidharan Nair, M.S. Meharban 
  Abstract: The effective training of deep architectures mainly depends on a large number of well-explained data. This is a problem in the medical field where it is hard and costly to gain such images. The tiny blood-vessels of the retina are the only part of the human structure that can be directly and nonintrusively foreseen within the living person. Hence, it can be easily obtained and examined by automatic tools. Fundus imaging is a basic check-up process in ophthalmology that provides essential data to make it easier for doctors to detect various eye-related diseases at early stages. Fundus image generation is a challenging process to carry out by constructing composite models of the eye structure. In this paper, we overcome the issue of unavailability of medical fundus datasets by synthesising them artificially through an encoder-decoder generator model to the existing MCML method of generative adversarial networks for easier, quicker, and early analysis.
  Keywords: fundus image; generative adversarial networks; encoder-decoder model; image synthesis; deep learning.
  
  
• Blockchain-based secure deduplication against duplicate-faking attack in decentralized storage  
  by Jingkai Zhou, Guohua Tian, Jianghong Wei 
  Abstract: Secure client-side deduplication enables cloud server to efficiently save storage space and communication bandwidth without compromising privacy. However, the potential duplicate-faking attack (DFA) may cause data users to lose their outsourced data. Existing solutions can either only detect DFA and fail to avoid data loss, or have high storage costs. In this paper, we propose a blockchain-based secure deduplication scheme against DFA in decentralised storage. Specifically, we firstly proposed a client-side deduplication protocol, in which server does not need to store additional metadata to check subsequent uploaders, who only need to encrypt the challenged partial blocks instead of the entire file. Besides, we design a battle mechanism based on smart contract to protect users from losing data. When an uploader detects a DFA, he can apply for a battle with the previous uploader to achieve an effective punishment. Finally, security and performance analysis indicate the practicality of the proposed scheme.
  Keywords: secure data deduplication; duplicate-faking attack; proof of ownership; blockchain; smart contract.
  
  
• Blockchain-based collaborative intrusion detection scheme  
  by Tianran Dang, Guohua Tian, Jianghong Wei, Shuqin Liu 
  Abstract: The collaborative intrusion detection technique is an effective solution to protect users from various cyber-attacks, among which the large-scale trusted sharing and real-time updating of attack instances are the main challenges. However, the existing collaborative intrusion detection systems (CIDS) either can only achieve real-time instance sharing in a centralised setting or implement large-scale instance sharing through blockchain. In this paper, we propose a novel blockchain-based CIDS scheme. Specifically, we present a reputation-based consensus protocol, which incentives service providers (SP) to evaluate the attack instances collected from collectors and punishes the malicious evaluators. Then, only trusted attack instances will be published on the blockchain to provide large-scale trusted intrusion detection services. Furthermore, we introduce a redactable blockchain technique to achieve dynamic instances update, which enables our scheme to provide a real-time intrusion detection service. Finally, we demonstrate the practicality of the proposed scheme through security analysis, theoretical analysis and performance evaluation.
  Keywords: collaborative intrusion detection; blockchain; reputation-based consensus; redactable blockchain.
  
  
• Application of bagging and particle swarm optimisation techniques to predict technology sector stock prices in the era of the Covid-19 pandemic using the support vector regression method  
  by Heni Sulastri, Sheila Maulida Intani, Rianto Rianto 
  Abstract: The increase in positive cases of Covid-19 not only affects the health and lifestyle, but also the economy and the stock market. Tech and digital sector stocks can be predicted to be one of the most profitable. Therefore, the prediction of the stock price is required to be able to see how the prospects of investment in the future. In this study, the prediction of the stock prices of Multipolar Technologies Ltd (MLPT) was carried out using the Support Vector Regression (SVR) method with Bootstrap Aggregation (Bagging) Technique and Particle Swarm Optimisation (PSO) as SVR optimisation. From the results of the prediction process, it is shown that the application of bagging and PSO techniques in predicting stock prices in the technological sector can reduce the Root Mean Squared Error (RMSE) value on the SVR, the RMSE value from 22.142 to 21.833. Although it does not have a big impact, it is better to apply a combination of bagging and PSO techniques to SVR than one of them (SVR / SVR - PSO / SVR-bagging).
  Keywords: bootstraps aggregation; bagging; Covid-19; particle swarm optimisation; prediction; share prices; support vector regression.
  
  
• Kidney diseases classification based on SONN and MLP-GA in ultrasound radiography images  
  by Anuradha Laishram, Khelchandra Thongam 
  Abstract: A strategy for robust classification of renal ultrasound images for the identification of three kidney disorders, renal calculus, cortical cyst, and hydronephrosis, has been attempted. Features were retrieved using the Intensity Histogram (IH), grey level co-occurrence matrices (GLCM), and grey level run length matrices (GLRLM) techniques. Using the extracted features, input samples are created and then fed to a hybrid model which is a combination of self-organising neural network (SONN) and multilayer perceptron (MLP) trained with a genetic algorithm (GA). Self-Organising Neural Network (SONN) is used to cluster the input patterns into four groups or clusters and finally, MLP using genetic algorithm is employed on each cluster to classify the input patterns. The proposed hybrid method using SONN and MLP-GA has more potential to classify the ultrasound images by achieving a precision of 93.9%, recall of 93.0%, F1 score of 93.0%, and overall accuracy of 96.8%.
  Keywords: genetic algorithm; grey level co-occurrence matrix; grey level run length matrix; intensity histogram; multilayer perceptron; self-organising neural network; ultrasound images.
  
  
• LightNet: pruned sparsed convolutional neural network for image classification  
  by Edna Too 
  Abstract: Deep learning has become a most sought after approach in the area of Artificial intelligence (AI). However, deep learning models pose some challenges in the learning process. It is computationally intensive to train deep learning networks and also resource intensive. Therefore, it cannot be applied in limited resource devices. Limited research is being done on implementation of efficient approaches for real world problems. This study tries bridge the gaps towards an applicable system in real world, especially in the agricultural sector for plant disease management and fruit classification. We introduce a novel architecture called LightNet. LighNet is an architecture that employs two strategies to achieve sparsity of DenseNet: the skip connections and pruning strategy. The resultant is a small network with reduced parameters and model size. Experimental evaluation reveal that LightNet is more efficient that the DenseNet architecture. The model is evaluated on real world datasets PlantsVillage and Fruits-360.
  Keywords: image classification; deep learning; convolution neural network; LightNet; ConvNet.
  
  
• Predicting possible antiviral drugs against COVID-19 based on Laplacian regularised least squares and similarity kernel fusion  
  by Xiaojun Zhang, Lan Yang, Hongbo Zhou 
  Abstract: COVID-19 has produced a severe impact on global health and wealth. Drug repurposing strategies provide effective ways for inhibiting COVID-19. In this manuscript, a drug repositioning-based virus-drug association prediction method, VDA-LRLSSKF, was developed to screen potential antiviral compounds against COVID-19. First, association profile similarity matrices of viruses and drugs are computed. Second, similarity kernel fusion model is presented to combine biological similarity and association profile similarity from viruses and drugs. Finally, a Laplacian regularised least squares method is used to compute the probability of association between each virus-drug pair. We compared VDA-LRLSSKF with four of the best VDA prediction methods. The experimental results and analysis demonstrate that VDA-LRLSSKF calculated better AUCs of 0.8286, 0.8404, 0.8882 on three datasets, respectively. VDA-LRLSSKF predicted that ribavirin and remdesivir could be underlying therapeutic clues for inhibiting COVID-19 and need further experimental validation.
  Keywords: SARS-CoV-2; VDA-LRLSSKF; drug repurposing; Laplacian regularised least squares; similarity kernel fusion.
  
  
• A modified Brown and Gibson model for cloud service selection  
  by Munmun Saha, Sanjaya Kumar Panda, Suvasini Panigrahi 
  Abstract: Cloud computing has been widely accepted in the information technology (IT) industry as it provides on-demand services, lower operational and investment costs, scalability, and many more. Nowadays, small and medium enterprises (SMEs) use the cloud infrastructure for building their applications, which makes their business more agile by using elastic and flexible cloud services. Many cloud service providers (CSPs) have offered numerous services to their customers. However, owing to the vast availability of cloud services and the wide range of CSPs, decision-making for cloud selection or adopting cloud services is not consistently straightforward. This paper proposes a modified Brown and Gibson model (M-BGM) to select the best CSP. We consider both the subjective and objective criteria for the non-quantifiable and quantifiable values, respectively. Here, various decision-makers can express their views about the alternatives. We compare M-BGM with multi-attribute group decision-making (MAGDM) approach and perform a sensitivity analysis to show the robustness.
  Keywords: cloud computing; multi-criteria decision-making; quality of service; cloud service provider; Brown and Gibson model; analytic hierarchy process; Delphi method; decision maker.
  
  
• 3DL-PS: an image encryption technique using a 3D logistic map, hashing functions and pixel scrambling techniques  
  by Parth Kalkotwar, Rahil Kadakia, Ramchandra Mangrulkar 
  Abstract: With the advancement in technology over the years, the security of data transferred over the internet is a major concern. In this paper, a robust and efficient image encryption scheme has been implemented using a 3D logistic map, SHA-512, and pixel scrambling. A good image encryption scheme should be able to produce two drastically different encrypted images for two original images with minute differences. Chaotic systems have proved to be highly efficient in providing this property, mainly because of their high randomness and volatility depending on the initial conditions. A 3D logistic map can be preferred over a 1D Logistic map owing to its increased encryption complexity, enhanced security, and better chaotic properties. To start the process, two secret keys are generated using two different user-provided keys and the input image, which makes it resistant to classical attacks such as the chosen-plaintext attack and chosen ciphertext attack. Further, it is necessary to change the pixel values of the original image so that it becomes difficult to trace back the original image from the encrypted image. Pixels of the images are altered using the values obtained upon the iteration of the 3D logistic map. In addition to this, two different pixel scrambling techniques are employed to enhance security. Firstly, different fragments of varying sizes are swapped depending on the secret keys generated earlier. Finally, a jumbling technique is used to mix the pixels horizontally and vertically in a completely dynamic way depending on the secret keys. The keyspace of the algorithm is found to be large enough to resist brute force attacks. The encrypted image has been observed and analysed against several attacks such as classical attacks, statistical attacks, and noise resistance. Key sensitivity analysis has also been performed. The results prove that the 3DL-PS algorithm is found to be resistant to several well-known attacks, providing an efficient image encryption scheme that can be used in various real-time applications.
  Keywords: image cryptography; chaotic systems; pixel scrambling; SHA-512; security analysis; classical attacks; 3D logistic map; noise resistance.
  DOI: 10.1504/IJCSE.2022.10049693
   
  
• Hybrid grasshopper and ant lion algorithms to improve imperceptibility, robustness and convergence rate for the video steganography  
  by Sahil Gupta, Naresh Kumar Garg 
  Abstract: The need for securing multimedia content from being intercepted is a prominent research issue. This work proposes an optimised video steganography model that improves imperceptibility and robustness by extracting keyframes and calculating the optimal scaling factor. The Squirrel Search Algorithm (SSA) is used to extract keyframes since it ensures distinct position updation processes through Levy flying and predator features, whilst the grasshopper optimisation and ant lion optimisation algorithms are hybridised to compute the optimal value of the scaling factor. In terms of imperceptibility and robustness, the simulation results suggest that the proposed approach outperforms existing data-hiding models. It also discovers the optimal scaling factor in under 10 iterations, indicating that the fastest convergence rate is possible.
  Keywords: ant-lion optimisation; grasshopper optimisation; SVD; video steganography; impercepbility; robustness; PSNR; MSE.
  DOI: 10.1504/IJCSE.2022.10053218
   
  
• Human behaviour analysis based on spatio-temporal dual-stream heterogeneous convolutional neural network  
  by Qing Ye, Yuqi Zhao, Haoxin Zhong 
  Abstract: At present, there are still many problems to be solved in human behaviour analysis, such as insufficient use of behaviour characteristic information and slow operation rate. We propose a human behaviour analysis algorithm based on spatio-temporal dual-stream heterogeneous convolutional neural network (STDNet). The algorithm improves on the basic structure of the traditional dual-stream network. When extracting spatial information, the DenseNet uses a hierarchical connection method to construct a dense network to extract the spatial feature of the video RGB image. When extracting motion information, BNInception is used to extract temporal features of video optical flow images. Finally, feature fusion is carried out by multi-layer perceptron and sent to Softmax classifier for classification. Experimental results on the UCF101 data set show that the algorithm can effectively use the spatio-temporal feature information in video, reduce the amount of calculation of the network model, and greatly improve the ability to distinguish similar actions.
  Keywords: human behaviour analysis; STDNet; optical flow; feature extraction; dual-stream network.
  DOI: 10.1504/IJCSE.2022.10048568
   
  
• High-volume transaction processing in bitcoin lightning network on blockchains  
  by Rashmi P. Sarode, Divij Singh, Yutaka Watanobe, Subhash Bhalla 
  Abstract: Transactions on e-commerce platforms using blockchain technology are required to face a high volume of executing transactions. These systems are required to be scalable. Bitcoin Lightning Network (BLN) can execute high volumes of transactions and is scalable due to few hops in the network. It is an off-chain payment channel built on top of a blockchain which speeds up the transactions. In this paper, we discuss BLN along with its transaction processing in detail, its benefits and applications. Additionally, we discuss alternative networks for payments. We also propose a secure model on BLN that can be used for any e-commerce platform and compare it with existing applications such as that of Ethereum and Stellar.
  Keywords: lightning network; bitcoin; cryptocurrency; blockchain; Ethereum.
  
  
• Data augmentation using fast converging CIELAB-GAN for efficient deep learning dataset generation  
  by Amin Fadaeddini, Babak Majidi, Alireza Souri, Mohammad Eshghi 
  Abstract: The commercial deep learning applications require large training datasets with many samples from different classes. The Generative Adversarial Networks (GAN) are able to create new data samples for training these machine learning models. However, the low speed of training these GANs in image and multimedia applications is a major constraint. In order to address this problem, in this paper a fast converging GAN called CIELAB-GAN for synthesizing new data samples for image data augmentation is proposed. The CIELAB-GAN simplifies the training process of GANs by transforming the images to the CIELAB colour space with fewer parameters. Then, the CIELAB-GAN translates the generated greyscale images into colourized samples using an autoencoder. The experimental results show that the CIELAB-GAN has lower computational complexity of 20% compared to the state of the art GAN models and is able to be trained substantially faster. The proposed CIELAB-GAN can be used for generating new image samples for various deep learning applications.
  Keywords: generative adversarial networks; deep learning; data augmentation; image processing.
  
  
• Aerial remote sensing image registration based on dense residual network of asymmetric convolution  
  by Ying Chen, Wencheng Zhang, Wei Wang, Jiahao Wang, Xianjing Li, Qi Zhang, Yanjiao Shi 
  Abstract: The existing image registration frameworks pay less attention to important local feature information and part of global feature information, resulting in low registration accuracy. However, asymmetric convolution and dense connection can pay more attention to the key information and shallow information of the image. Therefore, this paper propose a novel feature extraction module to improve the feature extraction ability and registration accuracy of the model. Asymmetric convolution and dense connection are used to improve the residual structure to focus on both local and global information in the feature extraction stage. In the feature matching stage, bidirectional matching is used to alleviate asymmetric matching results by fusing two outcomes. Furthermore, a secondary affine transformation is proposed to estimate the real transformation between two images adequately. In contrast with several popular algorithms, the proposed method has a better registration effect on two public datasets, which has practical significance.
  Keywords: remote sensing image registration; residual network; asymmetric convolution; dense connection; transfer learning; regularization; affine transformation.
  
  
• Non-parametric combination forecasting methods with application to GDP forecasting  
  by Wei Li, Yunyan Wang 
  Abstract: This work is devoted to constructing non-parametric combination prediction method, which can improve the forecasting effect and accuracy to some extent. In this paper, in order to forecast the regional gross domestic product, non-parametric autoregressive method is introduced into the autoregressive integrated moving average model, and a combined method of ARIMA model and non-parametric autoregressive model is established based on the residual correction. Furthermore, the specific prediction steps are proposed. The empirical results show that the new proposed combined model outperforms both the ARIMA model and the non-parametric autoregressive model in terms of regression effect and forecasting accuracy. The combination of parametric model and non-parametric model not only provides a method with better applicability and prediction effect for the establishment of GDP prediction model, but also provides a theoretical basis for the prediction of relevant economic data in the future. The prediction results show that during the Chinas 14th Five-Year Plan period, the gross domestic product of Jiangxi Province will increase by 7.01% annually.
  Keywords: GDP; ARIMA model; non-parametric autoregressive model; residual correction; combined model.
  
  
• Comparative study of point matching method with spectral method on numerical solution electromagnetic problems  
  by Mahmoud Behroozifar 
  Abstract: The present study focuses on comparing the point matching method and spectral method for solving the integral equations arising in the electromagnetic domain. The point matching method, which is a traditional method, was based on basis functions which most of the time results in a singular and ill-posed system of nonlinear equations. In order to prevent these inconveniences, the physical structure of the object must be altered in some cases this yields a high error in the results and requires high CPU time and memory usage. Also in most cases, this method converges slowly and leads a singular and ill-posed system. Consequently, applying the point matching method for this problem causes to obtain an approximate solution with low accuracy and high computation volume. As an alternative, we present the spectral method based on Bernstein polynomials (BPs) as a robust nominee. Employing the BPs reduces the problem to an algebraic equations system. The other merits of the presented method are faster convergence and avoidance of occurring a singular system.
  Keywords: Bernstein polynomials; electrostatic; micro strip; point matching method; spectral method.
  
  
• RCRE: radical-aware causal relationship extraction model oriented in the medical field  
  by Xiaoqing Li, Guangli Zhu, Zhongliang Wei, Shunxiang Zhang 
  Abstract: In the massive medical texts, the accuracy of causal relationship extraction is relatively low because of its special characteristic, a high correlation between semantics and radicals. To improve the extraction accuracy, this paper proposes a radical-aware causal relationship extraction model, which is oriented to the medical field. The BERT pre-training model is used to extract character-level features, which contain rich context information. To further deeply capture the semantics of characters, the Word2Vec model is used to extract radical features. Finally, the above two features are concatenated and passed into the extraction model to obtain the extraction results. Experimental results show that the proposed model can improve the accuracy of causal relationship extraction in medical texts.
  Keywords: causal relationship extraction; the medical field; radical features; BERT model; Word2Vec model.
  DOI: 10.1504/IJCSE.2023.10054227
   
  
• Detection of computationally intensive functions in a medical image segmentation algorithm based on an active contour model  
  by Carlos Gulo, Antonio Sementille, João Tavares 
  Abstract: Common image segmentation methods are computationally expensive, particularly when run on large medical datasets, and require powerful hardware to achieve image-based diagnosis in real-time. For a medical image segmentation algorithm that is based on an active contour model, our work presents an efficient approach that detects computationally intensive functions and adapts the implementation for improved performance. We employ profiling methods that assess algorithm performance taking into account the overall cost of execution, including time, memory access, and performance bottlenecks. We apply performance analysis techniques commonly available in traditional computing operating systems, which obviates the need for new setup or measurement techniques ensuring a short learning curve. The article presents guidelines to aid researchers in a) using profiling tools and b) detecting and checking potential optimisation snippets in medical image segmentation algorithms by measuring overall performance bottlenecks.
  Keywords: medical image processing and analysis; profiling tools; performance analysis; high-performance computing.
  DOI: 10.1504/IJCSE.2022.10050929
   
  
• Spaced retrieval therapy mobile application for Alzheimer's patients: a usability testing  
  by Kholoud Aljedaani, Reem Alnanih 
  Abstract: Alzheimers disease is the most common type of dementia. Statistics predict a sharp increase in patient numbers by 2050. Many applications support the patients in their daily activities and help them to engage in society. However, designing an acceptable and usable interface for this type of user is challenging. Spaced Retrieval Therapy (SRT) is a non-pharmacological therapy for Alzheimers disease that helps reduce the high cost of the treatments. The SRT application helps the patients to remember their vital information after a few sessions. In this paper, the authors develop the proposed application, which applies a non-pharmacological therapy to reduce the cost of treatments and help Alzheimers patients engage in society. The paper presents the findings on its usability. The usability test included 20 older adults divided into two groups (10 healthy and 10 with Alzheimers). Each group comprises two smaller groups (5 for each) to test the two types of interface. A list of tasks was given to both groups during the test, and the attributes of the task times and error numbers were collected. A post-task questionnaire evaluated the level of difficulty for each task. The result of the tasks confirmed that the Alzheimers group needed more time to complete the tasks than the healthy elderly group. Based on the post-task questionnaire, the healthy elderly group finds the default user interface simpler than the adapted one, which contrasts with the Alzheimers patients. Alzheimers patients performed faster in the adapted user interface. As recommendations: 1) use voice recognition instead of typing on keyboards because the typing tasks take the longest time in observation, and some Alzheimers patients cannot complete the tasks although they can read and write; 2) thicken the items borders in the menu because most errors result from confusion between the items.
  Keywords: spaced retrieval therapy; Alzheimer patients; usability testing; mobile application; designing user interface.
  DOI: 10.1504/IJCSE.2022.10050973
   
  
• Design of heuristic model to improve block-chain-based sidechain configuration  
  by Nisha Balani, Pallavi V. Chavan 
  Abstract: Data security is a major concern for any modern-day network deployment. Blockchain resolves security issues to a large extent. Blockchains are nowadays widely accepted for secure transactions and network communications. Since there is no limitation on the amount of data being stored, blockchain-based networks tend to become slow as the length of the main blockchain increases. To overcome this issue, the concept of sidechain is introduced. With sidechains, blockchain systems become faster, and inherit characteristics of blockchain including security, transparency and traceability. This paper proposes a solution for creating context-aware sidechains to increase system performance using a heuristic approach. The proposed algorithm assists in creation of customised sidechains via optimisation of blockchain mining delay using stochastic modelling. It generates a large number of stochastic sidechain combinations, evaluates them on the basis of mining delay, and selects optimal configuration. The proposed model is evaluated on different network conditions by varying network size and traffic density.
  Keywords: blockchain; sidechain; data sharing; fitness; QoS; blockchain mining delay; computation.
  DOI: 10.1504/IJCSE.2022.10050704
   
  
• Joint optimisation of feature selection and SVM parameters based on an improved fireworks algorithm  
  by Xiaoning Shen, Jiyong Xu, Mingjian Mao, Jiaqi Lu, Liyan Song, Qian Wang 
  Abstract: In order to reduce the redundant features and improve the accuracy in classification, an improved fireworks algorithm for joint optimisation of feature selection and SVM parameters is proposed. A new fitness evaluation method is designed, which can adjust the punishment degree adaptively with the increase of the number of selected features. A differential mutation operator is introduced to enhance the information interaction among fireworks and improve the local search ability of the fireworks algorithm. A fitness-based roulette wheel selection strategy is proposed to reduce the computational complexity of the selection operator. Three groups of comparisons on 14 UCI classification data sets with increasing scales validate the effectiveness of our strategies and the significance of joint optimisation. Experimental results show that the proposed algorithm can obtain a higher accuracy in classification with fewer features.
  Keywords: fireworks algorithm; support vector machines; feature selection; parameter optimisation; joint optimisation.
  
  
• Statistical analysis for predicting residents travel mode based on random forest  
  by Lei Chen, Zhengyan Sun, Shunxiang Zhang, Guangli Zhu, Subo Wei 
  Abstract: Random forest has achieved good results in the prediction task, but due to the complexity of travel mode and the uncertainty of random forest, the prediction accuracy of travel mode is low. To improve the accuracy of prediction, this paper proposes a residents travel modes prediction method based on the random forest. To extract valuable feature information, the questionnaire survey data is collected, which is preprocessed by three kinds of appropriate methods. Then, each feature is analysed by the statistical learning method to obtain the important feature of transportation selection. Finally, a random forest is constructed to predict the travel mode of residents selection of transportation. The parameters of random forests are modified and improved to achieve higher prediction accuracy of travel mode. The experimental results show that the method proposed in this paper effectively improves the prediction accuracy of the travel mode.
  Keywords: random forest; residents’ travel mode; statistical analysis.
  
  
• Wireless optimisation positioning algorithm with the support of node deployment  
  by Xudong Yang, Chengming Luo, Luxue Wang, Hao Liu, Lingli Zhang 
  Abstract: Position is one of the basic attributes of an object, which is one of the key technologies for its collaborative operation. As a distributed sensing method, Wireless Sensor Networks (WSNs) have become a feasible solution especially in satellite signal denied environments. Considering that the node deployment is the basis of target positioning in WSNs, this paper first researches the optimal deployment of wireless nodes, and then researches the optimal positioning of mobile targets. Based on the least squares equation, a feature matrix that can characterise the positioning error is derived so that the positioning error caused by wireless node deployment is minimised. Following that, the positioning results are refined using particle swarm optimisation, which makes the mobile target have a coarse to fine accuracy. The results indicate that the proposed algorithm can reduce the influence of network topology on positioning error, which is critical for some location-based applications.
  Keywords: distributed sensing; wireless positioning; node deployment; matrix eigenvalues; particle swarm.
  
  
• CNN-based battlefield classification and camouflage texture generation for real environments  
  by Sachi Choudhary, Rashmi Sharma 
  Abstract: It is critical to understand the environment in which the military forces are deployed. For self-defence and greater concealment, they should camouflage themselves. Camouflage is being used by the defence system to hide its personnel and equipment. The industry demands an intelligent system that can categorise the battlefield before generating texture for camouflaging their assets and objects, allowing them to adopt the conspicuous features of the scene. In this study, a CNN-based battlefield classification model has been developed to learn background information and classify the terrain. The study also intended to develop the texture for specific terrain by matching its salient features and boosting the effectiveness of the camouflage. Saliency maps have been used to measure the effectiveness of blending a camouflaged object into an environment.
  Keywords: digital camouflage; terrain classification; battlefield classification; camouflage generation; scene classification; colour clustering; saliency map.
  DOI: 10.1504/IJCSE.2022.10051287
   
  
• Investigation on the optimisation of Cholesky decomposition algorithm based on SIMD-DSP  
  by Huixiang Li, Huifu Zhang, Anxing Xie, Yonghua Hu, Wei Liang 
  Abstract: With the development of high-performance SIMD-DSP processors, corresponding highly efficient algorithms for matrix decomposition play an important role in the hardware performance of such processors. Cholesky decomposition is a fast decomposition method for symmetric positive definite matrices, which is widely used in matrix inversion and linear equation solving. According to the hardware characteristics of the FT-M7002 processors, in this paper, we optimise the algorithm in several ways. If the hardware has on-chip double-buffered memory, the parallel process of DMA transmitting and calculating is specially designed, which can hide most of the time cost of data movement and further improve the algorithms performance. The experimental results based on the FT-M7002 processor show that the performance of the optimised algorithm is 3.8~5.64 times that of the serial algorithm, and 1.39~2.14 times that of the TI library function.
  Keywords: Cholesky decomposition; DSP; SIMD.
  
  
• JALNet: joint attention learning network for RGB-D salient object detection  
  by Xiuju Gao, Jianhua Cui, Jin Meng, Huaizhong Shi, Songsong Duan, Chenxing Xia 
  Abstract: The existing RGB-D saliency object detection (SOD) methods mostly explore the complementary information between depth features and RGB features. However, these methods ignore the bi-directional complementarity between RGB and depth features. From this view, we propose a joint attention learning network (JALNet) to learn the cross-modal mutual complementary effect between the RGB images and depth maps. Specifically, two joint attention learning networks are designed, namely, a cross modal joint attention fusion module (JAFM) and a joint attention enhance module (JAEM), respectively. The JAFM learns cross-modal complementary information from the RGB and depth features, which can strengthen the interaction of information and complementarity of useful information. At the same time, we utilize the JAEM to enlarge receptive field information to highlight salient objects. We conducted comprehensive experiments on four public datasets, which proved that the performance of our proposed JALNet outperforms 16 state-of-the-art (SOTA) RGB-D SOD methods.
  Keywords: salient object detection; depth map; bi-directional complementarity; cross-modal features.
  
  
• Classifying blockchain cybercriminal transactions using hyperparameter tuned supervised machine learning models  
  by Rohit Saxena, Deepak Arora, Vishal Nagar 
  Abstract: Bitcoin is a crypto asset with transactions recorded on a decentralised, publicly accessible ledger. The real-world identity of a Bitcoin blockchain owner is masked behind a pseudonym, known as an address. As a result, Bitcoin is widely thought to provide a high level of anonymity, which is one of the reasons for its widespread use in criminal operations such as ransomware attacks, gambling, etc. As a result, classification and prediction of diverse cybercriminal users' activities and addresses in the Bitcoin blockchain are demanded. This research presents a classification of Bitcoin blockchain user activities and addresses associated with illicit transactions using supervised machine learning (ML). The labelled dataset samples with user activities are prepared using the unlabelled dataset available at the Blockchair repository and labelled dataset at WalletExplorer and are trained using classification models from the Decision Trees, Ensemble, Bayesian, and Instance-based Learning families. For balancing the classes of the dataset, the weighted mean and synthetic minority principles have been employed. The models' cross-validation (CV) accuracy is assessed. Extra Trees emerged as the best classification model, whereas Gaussian Na
  Keywords: blockchain; Bitcoin; supervised machine learning; classification; GridSearchCV.
  
  
• An improved blind/referenceless image spatial quality evaluator algorithm for image quality assessment  
  by Xuesong Li, Jinfeng Pan, Jianrun Shang, Alireza Souri, Mingliang Gao 
  Abstract: Image quality assessment (IQA) methods are generally studied in the spatial or transform domain. Due to the BRISQUE algorithm evaluating the quality of an image only based on its natural scene statistics of the spatial domain, the frequency features that are extracted from the modulation transfer function (MTF) are applied to improve its performance. MTF is estimated based on the slanted-edge method. The two-dimensional grey fitting algorithm is utilized to estimate the edge slope more accurately. Then the three-order Fermi function is utilized to match the preliminary estimated edge spread function to reduce the aliasing influence on MTF estimation. The features such as crucial frequency and the MTF value at Nyquist frequency are calculated and adopted to the BRISQUE method to assess the image quality. Experimental results on the image quality assessment databases illustrated that the proposed method outperforms the BRISQUE method and some other common methods, based on the linear and nonlinear correlation between the image quality assessed by the methods and their subjective value.
  Keywords: image assessment; modulation transfer function; Fermi function; feature extraction.
  DOI: 10.1504/IJCSE.2022.10051266
   
  
• Simple and compact finite difference formulae using real and complex variables  
  by Yohei Nishidate 
  Abstract: A new set of compact finite difference formulae is derived by simple combinations of the real and the complex Taylor series expansions. The truncation error is fourth-order in derived formulae for approximating first to fourth-order derivatives. Although there exist complex stencil finite difference formulae with better truncation errors, our formulae are computationally cheaper, requiring only three points for first to third-order and four points for fourth-order derivatives. The derived formulae are experimented with for approximating derivatives of relatively simple and highly nonlinear functions used in other literature. Although the new formulae suffer the subtractive cancellation, it is demonstrated that the derived formulae outperform finite difference formulae of comparable computational costs for relatively large step sizes.
  Keywords: Taylor series expansion; approximation in the complex domain; finite difference methods; compact finite difference formula; numerical approximation.
  
  
• A generalised incomplete no-equilibria transformation method to construct a hidden multi-scroll system with no-equilibrium  
  by Lihong Tang, Zongmei He, Yanli Yao, Ce Yang 
  Abstract: At present, there is a lot of research on multi-scroll chaotic systems with equilibrium points. However, there are few studies on no-equilibrium multi-scroll chaotic systems. This paper proposes a generalised incomplete no-equilibrium transformation method to design no-equilibrium multi-scroll chaotic systems. Firstly, a no-equilibrium chaotic system is constructed by adopting the proposed method. Phase plots and Lyapunov exponents show that the constructed no-equilibrium chaotic system can generate hidden hyperchaotic attractors. Then, a no-equilibrium multi-scroll hyperchaotic system is realized by introducing multi-level logic pulse signals. Theoretical analysis and numerical simulation show that the designed no-equilibrium multi-scroll hyperchaotic system can generate hidden multidirectional multi-double-scroll attractors including 1-D, 2-D, and 3-D hidden multi-scroll hyperchaotic attractors. Finally, an analogue circuit of the no-equilibrium multi-scroll hyperchaotic system is implemented by using commercial electronic elements. Various typical hidden multi-scroll attractors are verified on MULTISIM platform.
  Keywords: no-equilibrium; hidden attractors; multi-scroll; multi-level pulse.
  
  
• Selection of the best hybrid spectral similarity measure for characterising marine oil spills from multi-platform hyperspectral datasets  
  by Deepthi Deepthi, Deepa Sankar, Tessamma Thomas 
  Abstract: Marine oil pollution causes major economic crises in major industrial sectors such as fishing, shipping and tourism. It affects marine life even decades afterwards, necessitating very quick detection and remediation. Unfortunately, from remote sensing Hyperspectral Images (HSI), it is very difficult to detect oils, as oil slicks and seawater have nearly similar spectral properties. Therefore a cohesive and synergistic Hybrid Spectral Similarity Measure (HSSM) evaluating the multi-class, multi-platform capability of hyperspectral marine oil spill image classification is identified and recommended in this paper. Hyperspectral Images (HSI) procured from Spaceborne (Earth Observation (EO-1) Hyperion) and Airborne (Airborne Visible/Infrared Imaging Spectrometer (AVIRIS)) platforms are employed here to discriminate marine spectral classes. The statistical parameters such as Overall Accuracy (OA), Kappa, ROC/PR curve, AUC/PRAUC, weighted Youden index (Jw), F1 score and noise performances provided crucial evidence to find the best HSSM and Spectral Information Divergence-Chi square distance (SID-CHI). The stochastic capabilities of SID in capturing spectrum variations among spectral bands and the robustness to noises, inherited from CHI, are significant for the improved accuracy attained by SID-CHI over other HSSM. From the observations, it is established that SID-CHI can be used as a novel method for the multi-class and multi-platform classification of marine oil spill hyperspectral datasets.
  Keywords: hybrid spectral similarity measure; hyperspectral image; ROC curve; weighted Youden index; F1 score; optimal cut-off value.
  DOI: 10.1504/IJCSE.2022.10056248
   
  
• SAPNN: self-adaptive probabilistic neural network for medical diagnosis  
  by Yibin Xiong, Jun Wu, Qian Wang, Dandan Wei 
  Abstract: Medical diagnosis has always been a hot topic of great concern in the medical field. For this purpose, a self-adaptive probabilistic neural network (SAPNN) is proposed in this paper. Firstly, a hybrid cuckoo search (HCS) algorithm is proposed. Secondly, HCS is used in probabilistic neural networks for adapting the smoothing factor parameters. In order to accurately evaluate SAPNN proposed in this paper, the disease data sets of breast cancer, diabetes and Parkinsons disease were used for testing. Finally, comparison with several other methods yielded that the accuracy of SAPNN was the best in all cases, where the accuracy was 97.51%, 96.53%, 75.74%, 96.61%; recall was 97.6%, 99.12%, 79.74%, 88.24%; specificity was 96.15%, 88.88%, 59.03%, 95.31%; the precision was 97.85%, 94.32%, 80.12%, 85%, respectively. The results of various evaluation indexes show that the proposed SAPNN in this paper is a new method that can be applied to medical diagnosis.
  Keywords: ancillary diagnosis of disease; cuckoo search; information sharing; mutation strategy; probabilistic neural network.
  DOI: 10.1504/IJCSE.2022.10053093
   
  
• Minimum redundancy maximum relevance and VNS-based gene selection for cancer classification in high-dimensional data  
  by Ahmed Bir-Jmel, Sidi Mohamed Douiri, Souad Elbernoussi 
  Abstract: DNA microarray is a technique for measuring the expression levels of a huge number of genes. These levels have a significant impact on cancer classification tasks. In DNA datasets, the number of genes exceeds the number of samples that make the presence of irrelevant or redundant genes possible, which penalises the performance of classifiers. For that, the development of new methods for gene selection represents an active subject for researchers. In this paper, two-hybrid multivariate filters for gene selection, named VNSMI and VNSCor, are presented. The two methods surpass the univariate filters by considering the possible interaction between genes through the search for an optimal subset of genes that contains the minimum redundancy and the maximum relevance (MRMR). In the first stage of our proposed methods, we use a univariate filter by selecting the best-ranked genes based on information theory and the Pearson Correlation Coefficient (PCC). Then, we apply the Variable Neighbourhood Search (VNS) metaheuristic coupled with an innovative Stochastic Local Search (SLS) algorithm to find the final subset of genes that maximise the MRMR objective function. Evaluating the proposed method, the experiments were performed on six well-replicated microarray datasets. The obtained results show that the proposed approach leads to encouraging results in terms of accuracy and the number of selected genes. Also, improvements are observed consistently using the classifiers 1NN and SVM.
  Keywords: gene selection; feature selection; cancer classification; VNS; stochastic local search; normalised mutual information; MRMR; DNA microarray.
  
  
• Synthesis and evaluation of the structure of CAM memory by QCA computing technique  
  by Nirupma Pathak, Neeraj Kumar Misra, Santosh Kumar 
  Abstract: The lithographically based CMOS technology revolutions of the past few years are long behind us, but the technology used in today's microelectronics faces significant challenges in terms of speed, area, and power consumption. In the domain of QCA, the purpose of this article is to design a novel CAM memory. This article deals with the compact structure of the novel CAM memory, which is based on GDI-CMOS and QCA technology, respectively. Compared with the recently reported designs in the literature, it is observed that the area, latency, majority gate, and cell count of the proposed CAM are decreased by more than 78.57%, 50%, 40%, and 67%, respectively. In addition to this, the clock delay of the CAM cell design is less than the other results that have been reported. This cutting-edge QCA-based CAM structure is not only one of a kind but also very cost-effective in today's nano-devices. This CAM design that has been proposed improves performance while also making the use of modern device development simpler and more cost-effective.
  Keywords: CAM memory; QCA; GDI-CMOS; nano-electronics.
  DOI: 10.1504/IJCSE.2022.10053117
   
  
• Canopy centre-based fuzzy C-means clustering for enhancement of soil fertility Prediction  
  by M. Sujatha, C.D. Jaidhar 
  Abstract: For plants to develop, fertile soil is necessary. Estimating soil parameters based on time change is crucial for enhancing soil fertility. Sentinel-2s remote sensing technology produces images that can be used to gauge soil parameters. In this study, values for soil parameters such as electrical conductivity, pH, organic carbon, and nitrogen are derived using Sentinel-2 data. In order to increase the clustering accuracy, this study suggests using canopy centre-based fuzzy C-means clustering and comparing it with manual labelling and other clustering techniques, such as canopy density-based, expectation maximisation, farthest-first, k-Means, and fuzzy C-means clustering. The proposed clustering achieved the highest clustering accuracy of 78.42%. Machine learning-based classifiers were applied to classify soil fertility, including naive Bayes, support vector machine, decision trees, and random forest (RF). A dataset labelled with the proposed RF clustering classifier achieves a high classification accuracy of 99.69% with 10-fold cross-validation.
  Keywords: clustering; classification; machine learning; remote sensing; soil fertility.
  
  
• Research on mobile robot path planning and tracking control  
  by Jieyun Yu 
  Abstract: Autonomous navigation of a robot is a promising research domain due to its extensive applications in which planning and motion control are the most important and interesting parts. The proposed techniques are classified into two main categories: the first session focus on the improvement model free adaptive control (MFAC) to meet the extreme performances of the control system, and the second concentrates on the classic artificial potential field (APF) algorithm to deal with the limitations like falling into local minima and a non-reachable goal problem. This paper proposes a novel exponential feedforward-feedback control strategy based on iterative learning control (ILC) MFAC to the reference trajectory tracking, and then introduces a virtual target with exponential coordinated form to realise local risk collision avoidance for path planning. Compared to some traditional models, our proposed methods have a faster trajectory convergence rate, lower avoidable error, and higher safe performance. The simulation results verify that our work would bring meaningful insights to future intelligent navigation research.
  Keywords: trajectory tracking; path planning; model-free adaptive control; artificial potential field; exponential-form virtual target.
  DOI: 10.1504/IJCSE.2023.10054169
   
  
• Texture-based superpixel segmentation algorithm for classification of hyperspectral images  
  by Subhashree Subudhi, Ramnarayan Patro, Pradyut Kumar Biswal 
  Abstract: To increase classification accuracy, a variety of feature extraction techniques have been presented. A preprocessing method called superpixel segmentation divides an image into meaningful sub-regions, which simplifies the image. This substantially reduces single-pixel misclassification. In this work, a texture-based superpixel segmentation technique is developed for the accurate classification of Hyperspectral Images (HSI). Initially, the local binary pattern and Gabor filters are employed to extract local and global image texture information. The extracted texture features are then provided as input to the Simple Linear Iterative Clustering (SLIC) algorithm for segmentation map generation. The final classification map is constructed by using a majority vote strategy between the superpixel segmentation map and the pixel-wise classification map. The proposed method was validated on standard HSI datasets. In terms of classification performance, it outperformed other state-of-the-art algorithms. Furthermore, the algorithm may be incorporated into the UAV's onboard camera to automatically classify HSI.
  Keywords: hyperspectral image classification; superpixel segmentation; SLIC; spatial-spectral feature extraction.
  
  
• FedCluster: a global user profile generation method based on vertical federated clustering  
  by Zheng Huo, Ping He, Lisha Hu 
  Abstract: Federated learning can serve as a basis to solve the data island problem and data privacy leakage problem in distributed machine learning. This paper proposes a privacy-preserving algorithm referred to as FedCluster, to construct a global user profile via vertical federated clustering. The traditional k-medoids algorithm was then extended to the federated learning architecture to construct the user profiles on vertical segmented data. The main interaction parameter between the participants and the server was the distance matrix from each point to the k medoids. Differential privacy was adopted to protect the privacy of the participant data during the exchange of training parameters. We conducted experiments on a real-world dataset. The results revealed that the precision of FedCluster reached 81.87%. The runtime exhibited a linear increase with an increase in the dataset size and the number of participants, which indicates a high performance in terms of precision and effectiveness.
  Keywords: federated learning; footrule distance; k-mediods clustering; order preserving encryption.
  
  
• Value chain for smart grid data: a brief review  
  by Feng Chen, Huan Xu, Jigang Zhang, Guiyu Li 
  Abstract: Smart grids are now crucial infrastructures in many countries. They build two-way communication between customers and utility enterprises. Since power and energy are associated with human activities, smart grid data are extremely valuable. At present, these data are currently being used in some areas. As a novel asset, the value of smart grid data needs quantitative measurement for evaluation and pricing. To achieve this, it is essential to analyse the overall process of value creation, which can help to calculate costs and discover potential applications. The process can be effectively revealed by building a data value chain for smart grid data, which illustrates the data flow and clarifies the data sources, analytics, utilization, and monetization. This article provides a three-step data value chain for smart grid data and expounds on each step. This article also reviews various methods and some challenges with smart grid data.
  Keywords: smart grid; data value chain; data collection; data analysis; data monetisation.
  
  
• A search pattern based on the repeated motion vector components for the fast block matching motion estimation in temporal coding  
  by Awanish Mishra, Narendra Kohli 
  Abstract: To reduce the amount of unnecessary data in a video's timeline, block-based motion estimate is routinely used. However, a significant reduction in the computational complexity of motion estimation remains a significant problem. In this paper, a search pattern approach is proposed to efficiently estimate the motion of blocks. The proposed algorithm estimates the motion based on the maximum frequency of magnitude and direction of the available motion vector components. Motion vector components with higher frequency have greater probability to provide early estimation of matching block. In this iterative process, searching for the matching block is terminated on getting the matched block. To demonstrate the enhanced performance of the proposed approach, a comprehensive analysis is carried out, and when the results are compared, the novel approach outperforms recent motion estimation approaches. The proposed approach improves the best case complexity until it finds one search per block for dynamic blocks. It improves the average case complexity because of the early termination of the process.
  Keywords: motion estimation; block matching; search parameter; source frame; reference frame.
  
  
• Pyramid hierarchical network for multispectral pan-sharpening  
  by Zenglu Li, Xiaoyu Guo, Songyang Xiang 
  Abstract: Pan-sharpening aims to fuse high spatial-resolution panchromatic images (PAN) and low spatial-resolution multispectral images (MS) into high spatial-resolution multispectral images (HRMS). We propose a pyramid hierarchical multi-spectral fusion network, called PH-Net, which can automatically fuse MS images and PAN images to generate corresponding HRMS images. The architecture is based on the U-Net network. First, a multi-level receptive field is realised by constructing an input pyramid. Then, hierarchical features are extracted from the encoder, decoder, and input pyramid. Finally, the rich hierarchical features are used to calculate the residual error between the MS image and the corresponding HRMS image. The learned residual error is inserted into the MS image to obtain the final high spatial-resolution multispectral image. To demonstrate the effectiveness of each component in the network architecture, we conducted an ablation study. In addition, thanks to the design of the multi-layer architecture, model training does not require a large dataset, which greatly improves the training speed and significantly improves the generalizability and ease of deployment of this work in the field of remote sensing images. Through qualitative and quantitative experiments, we proved that the proposed method is superior to current advanced methods.
  Keywords: pan-sharpening; image fusion; pyramid Attention; multispectral image; deep learning.
  DOI: 10.1504/IJCSE.2022.10053377
   
  
• Clustering ensemble by clustering selected weighted clusters  
  by Arko Banerjee, Suvendu Chandan Nayak, Chhabi Rani Panigrahi, Bibudhendu Pati 
  Abstract: Owing to the fact that no single clustering approach is capable of producing the optimal result for any given data, the notion of clustering ensembles has emerged, which attempts to extract a novel and robust consensus clustering from a given ensemble of base clusterings of the data. While forming the consensus, weights can be assigned to the base clusterings or their constituent clusters to prioritise those that accurately represent the underlying structure of the data. In this paper, we present a novel method of cluster selection from base clusterings and subsequently merging selected clusters into the desired number of clusters in order to build a high-quality consensus clustering without gaining access to the internal distribution of data points. The method has been shown to work well with a wide range of data and to be better than many well-known clustering methods.
  Keywords: clustering ensemble; weighted clustering; entropy; cluster selection.
  
  
• Cell counting via attentive recognition network  
  by Xiangyu Guo, Jinyong Chen, Guisheng Zhang, Guofeng Zou, Qilei Li, Mingliang Gao 
  Abstract: Accurate cell counting in biomedical images is a fundamental yet challenging task for disease diagnosis. The early manual cell counting methods are mainly based on detection and regression, which are time-consuming and prone to errors. Benefitting from the advent of deep learning, convolutional neural network (CNN)-based cell counting has become the mainstream method. Despite the outstanding performance of CNN-based cell counting methods, the complex tissue background in medical images still hinders the accuracy of cell counting. In this paper, to solve the problem of complex tissue background and improve the performance of cell counting, an attentive recognition network (ARNet) is built. Specifically, the ARNet is composed of five convolution blocks and a channel attention (CA) module. The convolution blocks are employed to extract the basic features, and the CA module is introduced to suppress the complex background by recalibrating the weight of each channel to pay more attention to cells. Subjective and objective experiments on synthetic bacterial cells dataset and modified bone marrow dataset prove that the proposed ARNet outperforms the mainstream methods in accuracy and stability.
  Keywords: healthcare; cell counting; attention mechanism; convolutional neural network.
  DOI: 10.1504/IJCSE.2022.10055133
   
  
• Information fusion and emergency knowledge graph construction of urban rail transit  
  by Guangyu Zhu, Rongzheng Yang, Jiaxin Fan, Wei Yun, Bo Wu, Qi Wu 
  Abstract: The core of an intelligent emergency system of urban rail transit (URT) is to build a knowledge system for operation and emergency management. This paper proposes a construction model of emergency knowledge graph of URT combined with information fusion. Firstly, the scheme layer of the knowledge graph is designed in the top-down style, which defines the knowledge framework, entity types and entity relationships of the knowledge graph. Secondly, the entity extraction model based on adversarial training and Bert is proposed to extract knowledge from the emergency record text. The information fusion method is used to normalise the knowledge extracted from multi-source data to complete the construction of the data layer. Finally, Neo4j graph database is used to store and manage the data, and then the emergency knowledge graph of URT is constructed. Experiments show that the extraction model proposed in this paper has a better extraction effect than the mainstream models in terms of F1 value, which is increased by 7.52%, 1.87% and 1.31%, respectively. In addition, the emergency knowledge graph of URT based on this method can better fuse multi-source information and provide better basic support for the construction of URT intelligent emergency system.
  Keywords: intelligent emergency; knowledge graph; knowledge extraction; information fusion; urban rail emergency knowledge graph.
  DOI: 10.1504/IJCSE.2022.10053044
   
  
• A bibliometric analysis of the application of deep learning in economics, econometrics, and finance  
  by Arash Salehpour, Karim Samadzamini 
  Abstract: This research looked at the deep learning applications in economics, econometrics, and finance. Two hundred and fifty articles from the Scopus database's index of journals published between 2013 and 2022 were gathered using a bibliometric technique. The data was analysed using many programs (R studio, Excel, and Biblioshiny), and in terms of countries, organisations, publications, papers, and authors, the most prominent scientific players were highlighted. Our research found that as of 2019, the quantity of publications has increased. The literature analysis received the most contributions from China and the United States. The most significant findings and discussions came from the following analyses: estimation of share prices, asset management price fluctuations and liquidity, forecast of bankruptcies, evaluation of credit risk, risk assessment, commodity prices top trend analysis, citation analysis, thematic evolution, and thematic map. Our findings offer practical recommendations on how deep learning may be implemented into decision-making processes for market participants, particularly those working in fintech and finance.
  Keywords: deep learning; bibliometrics; economics; econometrics; finance.
  
  
• Self-supervised learning with split batch repetition strategy for long-tail recognition  
  by Zhangze Liao, Liyan Ma, Xiangfeng Luo, Shaorong Xie 
  Abstract: Deep neural networks cannot be well applied to balance testing when the training data present a long tail distribution. Existing works improve the performance of the model in long tail recognition by changing the model training strategy, data expansion, and model structure optimisation. However, they tend to use supervised approaches when training the model representations, which makes the model difficult to learn the features of the tail classes. In this paper, we use self-supervised representation learning (SSRL) to enhance the model's representations and design a three-branch network to merge SSRL with decoupled learning. Each branch adopts different learning goals to enable the model to learn balanced image features in the long-tail data. In addition, we propose a Split Batch Repetition strategy for long-tailed datasets to improve the model. Our experiments on the Imbalance CIFAR-10, Imbalance CIFAR-100, and ImageNet-LT datasets outperform existing similar methods. The ablation experiments prove that our method performs better on more imbalanced datasets. All experiments demonstrate the effectiveness of incorporating the self-supervised representation learning model and split batch repetition strategy.
  Keywords: long-tail recognition; self-supervised learning; decoupled learning; image classification; deep learning; neural network; computer vision;.
  
  
• SLIC-SSA: an image segmentation method based on superpixel and sparrow search algorithm  
  by Hao Li, Hong Wen, Jia Li, Lijun Xiao 
  Abstract: Clustering algorithms are widely used in image segmentation owing to their universality. However, the methods based on clustering algorithms are sensitive to noise and easily fall into local optimum. To address these issues, we propose an image segmentation method (SLIC-SSA) based on superpixel method and sparrow search algorithm. Firstly, the presegmentation result is obtained by superpixel method. Owing to the use of local spatial information, the influence of noise can be reduced. Then, the clustering algorithm based on sparrow search algorithm is performed on superpixel image to complete the segmentation. To improve the quality of the results, the chaotic strategy is used to initialise the population. A fitness function is proposed to ensure the similarity within the cluster and the difference between the clusters. Experiments on real images show that the proposed method can obtain better results than comparative methods. Meanwhile, time consumption can be reduced.
  Keywords: clustering; image segmentation; sparrow search; superpixel; swarm intelligence optimisation.
  DOI: 10.1504/IJCSE.2023.10053888
   
  
• Title SMedia: social media data analysis for emergency detection and its type identification  
  by Sarmistha Nanda, Chhabi Rani Panigrahi, Bibudhendu Pati, Prasant Mohapatra 
  Abstract: Owing to the advancement of technology, social media can spread information very fast. People post information about themselves or about an event in the proximity of any emergency. However, proper analysis of social media data is necessary to address the challenges of emergency detection and its type identification. An early identification along with proper action is essential to minimize the loss due to occurrence of any type of emergency. In this work, authors used the keyword based tweets data to detect the emergency. First, the emergency tweets were classified using the proposed HDLed model and the accuracy obtained from the experimental study was 88% which was more as compared to the existing algorithms such as Convolutional Neural Network (CNN), Bidirectional-Long Short-Term Memory (Bi-LSTM), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU). Next, the type of emergency was detected using the baseline multiclass classifiers such as Na
  Keywords: social media; text classification; CNN; bidirectional LSTM; emergency.
  DOI: 10.1504/IJCSE.2023.10054663
   
  
• Integration of statistical parameters based color-texture descriptors for radar remote sensing image retrieval applications  
  by Naushad Varish, Sambidi Rohan Reddy, Nadimpalli Gautham Sashi Varma, Priyanka Singh 
  Abstract: In this paper, a novel image retrieval method based on colour-texture contents for radar remote sensing applications is proposed, where global properties-based colour contents are extracted from different number of groups of histograms of colour image planes and local properties-based texture contents have been derived from block level GLCM of an image plane. The integration of colour-texture contents represents the low dimensional feature which reduces overall computational overhead and increases the retrieval speed. To give importance to the feature components, suitable weights are imposed to both colour-texture contents appropriately. The obtained feature information is describing the radar image effectively and also similarity measures play a significant role for better performance. This work compares eight similarity metrics to select the best one in the retrieval process. To validate the suggested method, experiments on two image datasets are performed and decent retrieval results have been attained with rich colour-texture contents.
  Keywords: remote sensing image retrieval; statistical parameters; grey-level co-occurrence matrix; feature descriptors; min-max; similarity measures.
  
  
• Optimising group-by and aggregation on the coupled CPU-GPU architecture  
  by Hua Luan, Yan Fu 
  Abstract: The coupled CPU-GPU architecture as an emerging heterogeneous environment has attracted much attention from researchers. On this kind of architecture, the GPU is built on the same chip as the CPU. Different from the discrete GPU, there is no data transfer via the PCIe bus between the CPU and the integrated GPU, and the two processors share the same memory. Grouping and aggregation is an important and time-consuming operator in a DBMS. Whether the coupled GPU could be used to increase its performance is an interesting problem. In this paper, we study how to optimise grouping and aggregation based on chained hashing on the coupled CPU-GPU architecture. Two flexible co-processing strategies are proposed to take advantage of the hybrid computing resources effectively. A thorough set of experiments are conducted and the results show that the coupled GPU could help obtain better performance for group-by and aggregation.
  Keywords: hash grouping; coupled CPU-GPU architecture; co-processing.
  
  
• Distilling object detectors with mask-guided feature and relation-based knowledge  
  by Liang Zeng, Xiangfeng Luo, Liyan Liyan Ma, Yinsai Guo, Xue Chen 
  Abstract: Knowledge distillation(KD) is an effective technique for network compression and model accuracy enhancement in image classification, semantic segmentation, pre-trained language model, and so on. However, existing KD methods are specialized for image classification and cannot be used effectively for object detection tasks, with the following two limitations: the imbalance of foreground and background instances and the neglect distillation of relation-based knowledge. In this paper, we present a general Mask-guided Feature and Relation-based Knowledge distillation framework (MAR) consisting of two components, mask-guided distillation, and relation-based distillation, to address the above problems. The mask-guided distillation is designed to emphasize students' learning of close-to-object features via multi-value masks, while relation-based distillation is proposed to mimic the relational information between different feature pixels on the classification head. Extensive experiments show that our methods achieve excellent AP improvements on both one-stage and two-stage detectors. Specifically, Faster RCNN with ResNet50 backbone achieves 40.6% in mAP under 1x schedule on the COCO dataset, which is 3.2% higher than the baseline and even surpasses the teacher detector.
  Keywords: knowledge distillation; multi-value mask; object detection.
  
  
• Study on the capacity of a hybrid solar PV/wind turbine system using small-scale prototype application for dairy farm power demand in North Texas  
  by Dakota Messer, Hoe-Gil Lee 
  Abstract: Renewable energy systems are increasing in power production and efficiency, as the need for clean energy sources continues to increase. The purpose of this research is to design and analyse an integrated hybrid system that will provide a continuous power supply. A local dairy farms cattle cooling system operates at a power demand of 38,664.00 watts to 128,880.00 watts per day and is chosen as the load that is to be supported for an off-grid application. Approaching this design, the renewable technologies and specifications best suited for the location are chosen to include a horizontal axis wind turbine and a single axis solar panel with active tracking. A 120 kilowatt wind turbine and a 150 kilowatt solar panel kit is determined to be optimal through various analyses. Integrating the subsystems with energy conversion and battery storage systems provides the power demand at all times throughout the day as well as the effectiveness of the system. An economic analysis is completed to determine an investment recuperation within 8 years and estimated savings of $890,982.52 for the remainder of the system's life expectancy.
  Keywords: wind turbine; solar PV energy; dairy farm; hybrid renewable system.
  
  
• Examining the role of likes in follower network evolution based on a dynamic panel data model  
  by Tao Wang, Shuang Fu, Zhiyi Wu 
  Abstract: Posting product recommendation articles by content creators from the consumer group in social shopping communities has become an effective way to connect consumers to products. Content creators with larger follower counts have higher levels of influence. However, little is known about the causes of the evolution of their follower networks. Therefore, we examined the impact of social media likes that content creators received on the follower count and the moderating effect of the previous follower count on the role of likes. We achieve that by crawling real data from China's leading social shopping community. We empirically tested a dynamic panel data model and found that more likes are positively associated with the growth of the follower network size, while the previous follower count negatively moderates this effect. These findings have implications for researchers seeking to understand the antecedents of follower network evolution and for practitioners seeking to attract more followers.
  Keywords: content creator; social media like; follower network evolution; social shopping community; dynamic panel data model.
  DOI: 10.1504/IJCSE.2023.10055525
   
  
• WSN distributed sensing of mobile robot under the irregular topology of underground pipe gallery  
  by Gaifang Xin, Jun Zhu, Jing Tang 
  Abstract: As a long and narrow irregular structure, the underground pipe gallery can reach tens of kilometres. In order to avoid the tedious work by manual handheld or complex installation by wired networks, Wireless sensor networks (WSNs) integrated with mobile robot can be used for monitoring. It is worth noting that mobile robot needs to mark its corresponding position when collecting the WSNs data. Hence, this paper proposes a WSNs distributed sensing model for mobile robots. Within the communicable radius, the anchor node set for distributed sensing is firstly solved by combining the motion characteristics of mobile robot. Then dual mapping between wireless parameters and robot positions is established in irregular topology.
  Keywords: underground pipe gallery; mobile robot; distributed sensing; irregular structure; accuracy evaluation.
  DOI: 10.1504/IJCSE.2023.10055745
   
  
• An energy efficient on-demand multi-path routing protocol for wireless body area network  
  by Qingling Liu, Qi Wang 
  Abstract: Wireless body area network (WBAN) makes a remarkable contribution to healthcare. However, WBAN is faced with problems such as energy shortage and radiation safety. The existing researches on WBAN routing protocol tend to focus on energy saving, but ignore other problems. Therefore, in this paper, an energy efficient on-demand multi-path routing protocol for WBAN is proposed to prolong the network lifetime while ensuring the user’s radiation security and quality of service (QoS). Firstly, a cost function is constructed, which linearly combines the node residual energy ratio, the specific absorption rate (SAR), the buffer remaining space ratio and the node movement metric parameters. Secondly, the analytic hierarchy process (AHP) is adopted to calculate the weight vector of the cost function. Thirdly, the standard ad hoc on-demand multi-path distance vector protocol (AOMDV) is used as the base routing protocol and its routing mechanism is tuned to meet WBAN performance requirements using the cost function provided. Simulation results show that the revised routing protocol has better performance in terms of network throughput, average end-to-end delay and energy efficiency in a variety of simulation scenarios.
  Keywords: wireless body area network; WBAN; quality of service; QoS; specific absorption rate; SAR; analytic hierarchy process; AHP.
  DOI: 10.1504/IJCSE.2023.10055920
   
  

Special Issue on: CCPI'20 Smart Cloud Applications, Services and Technologies

• A big data and cloud computing model architecture for a multi-class travel demand estimation through traffic measures: a real case application in Italy  
  by Armando Cartenì, Ilaria Henke, Assunta Errico, Marida Di Bartolomeo 
  Abstract: The big data and cloud computing are an extraordinary opportunity to implement multipurpose smart applications for the management and the control of transport systems. The aim of this paper is to propose a big data and cloud computing model architecture for a multi-class origin-destination demand estimation based on the application of a bi-level transport algorithm using traffic counts on a congested network, also to propose sustainable policies at urban scale. The proposed methodology has been applied to a real case study in terms of travel demand estimation within the city of Naples (Italy), also aiming to verify the effectiveness of a sustainable policy in term of reducing traffic congestion by about 20% through en-route travel information. The obtained results, although preliminary, suggest the usefulness of the proposed methodology in terms of ability in real time/pre-fixed time periods to estimate traffic demand.
  Keywords: cloud computing; big data; virtualisation; smart city; internet of things; transportation planning; demand estimation; sustainable mobility; simulation model.
  
  
• A methodology for introducing an energy-efficient component within the rail infrastructure access charges in Italy  
  by Marilisa Botte, Ilaria Tufano, Luca D'Acierno 
  Abstract: After the separation of rail infrastructure managers from rail service operators occurred within the European Union in 1991, the necessity of defining an access charge framework for ensuring non-discriminatory access to the rail market arose. Basically, it has to guarantee an economic balance for infrastructure manager accounts. Currently, in the Italian context, access charge schemes neglect the actual energy-consumption of rail operators and related costs of energy traction for infrastructure managers. Therefore, we propose a methodology, integrating cloud-based tasks and simulation tools, for including such an aspect within the infrastructure toll, thus making the system more sustainable. Finally, to show the feasibility of the proposed approach, it has been applied to an Italian real rail context, i.e. the Rome-Naples high-speed railway line. Results have shown that customising the tool access charges, by considering the power supply required, may generate a virtuous loop with an increase in energy-efficiency of rail systems.
  Keywords: cloud-based applications; rail infrastructure access charges; environmental component; energy-saving policies.
  
  
• Edge analytics on resource-constrained devices  
  by Sean Savitz, Charith Perera, Omer Rana 
  Abstract: Video and image cameras have become an important type of sensor within the Internet of Things (IoT) sensing ecosystem. Camera sensors can measure our environment at high precision, providing the basis for detecting more complex phenomena in comparison with other sensors e.g. temperature or humidity. This comes at a high computational cost on the CPU, memory and storage resources, and requires consideration of various deployment constraints, such as lighting and height of camera placement. Using benchmarks, this work evaluates object classification on resource-constrained devices, focusing on video feeds from IoT cameras. The models that have been used in this research include MobileNetV1, MobileNetV2 and Faster R-CNN, which can be combined with regression models for precise object localisation. We compare the models by using their accuracy for classifying objects and the demand they impose on the computational resources of a Raspberry Pi.
  Keywords: internet of things; edge computing; edge analytics; resource-constrained devices; camera sensing; deep learning; object detection.
  
  
• Traffic control strategies based on internet of vehicles architectures for smart traffic management: centralised vs decentralised approach  
  by Houda Oulha, Roberta Di Pace, Rachid Ouafi, Stefano De Luca 
  Abstract: In order to reduce traffic congestion, real-time traffic control is one of the most widely adopted strategies. However, the effectiveness of this approach is constrained not only by the adopted framework but also by data. Indeed, the computational complexity may significantly affect this kind of application, thus the trade-off between the effectiveness and the efficiency must be analysed. In this context, the most appropriate traffic control strategy to be adopted must be accurately evaluated. In general, there are three main control approaches in the literature: centralised control, decentralised control and distributed control, which is an intermediate approach. In this paper, the effectiveness of a centralised and a decentralised approach is compared and applied to two network layouts. The results, evaluated not only in terms of performance index with reference to the network total delay but also in terms of emissions and fuel consumption, highlight that the considered centralised approach outperforms the adopted decentralised one and this is particularly evident in the case of more complex layouts.
  Keywords: cloud computing; internet of vehicles; transportation; centralised control; decentralised control; emissions; fuel consumption.
  
  
• ACSmI: a solution to address the challenges of cloud services federation and monitoring towards the cloud continuum  
  by Juncal Alonso, Maider Huarte, Leire Orue-Echevarria 
  Abstract: The evolution of cloud computing has changed the way in which cloud service providers offer their services and how cloud customers consume them, moving towards the usage of multiple cloud services, in what is called multi-cloud. Multi-cloud is gaining interest by the expansion of IoT, edge computing and the cloud continuum, where distributed cloud federation models are necessary for effective application deployment and operation. This work presents ACSmI (Advanced Cloud Service Meta-Intermediator), a solution that implements a cloud federation, supporting the seamless brokerage of cloud services. Technical details addressing the discovered shortcomings are presented, including a proof of concept built on JHipster, Java, InfluxD, Telegraf and Grafana. ACSmI contributes to relevant elements of the European Gaia-X initiative, specifically to the federated catalogue, continuous monitoring, and certification of services. The experiments show that the proposed solution effectively saves up to 75% of the DevOps teams effort to discover, contract and monitor cloud services.
  Keywords: cloud service broker; cloud services federation; cloud services brokerage; cloud services intermediation; hybrid cloud; cloud service monitoring; multi-cloud; DevOps; cloud service level agreement; cloud service discovery; multi-cloud service management; cloud continuum.
  
  
• User perception and economic analysis of an e-mobility service: development of an electric bus service in Naples, Italy  
  by Ilaria Henke, Assunta Errico, Luigi Di Francesco 
  Abstract: Among the sustainable mobility policies, electric mobility seems to be one of the best choices to reach sustainable goals, but it has limits that could be partially exceeded in the local public transport. This research presents a methodology to design a new sustainable public transport service that meets users needs by analysing economic feasibility. This methodology is then applied to a real case study: renewing an 'old' bus fleet with an electric one charged by a photovoltaic system in the city of Naples (Southern Italy). Its effects on users' mobility choices were assessed through a mobility survey. The bus line and the photovoltaic system were designed. Finally, the economic feasibility of the project was assessed through a cost-benefit analysis. This research is placed in the field of smart mobility and new technologies that increasingly need to store, manage, and process large amounts of data typical of cloud computing and big data applications
  Keywords: e-mobility; electric bus services; cloud computing; user perception; economic analysis; cost-benefit analysis; photovoltaic system; sustainable mobility policies; sustainable goals; new technologies; local emissions; environmental impacts.
  
  

Special Issue on: ICNC-FSKD 2021 Cutting-edge High-Performance Computing and Artificial Intelligence Technologies for Medical E-Diagnosis

• Research on tracking of moving objects based on depth feature detection  
  by Guocai Zuo, Xiaoli Zhang, Jing Zheng 
  Abstract: In complex conditions such as illumination change, target rotation, and background clutter, tracking drift or failure of target tracking may occur. Convolutional neural networks (CNN) can achieve robust target tracking in complex scenes such as illumination, rotation, background clutter, and so on. Therefore, this paper proposes a target tracking algorithm CNNT based on a convolutional neural network. Use CNN deep learning model to extract the deep features of the sample to complete the target detection task, and then use the Kernel Correlation Filter (KCF) target tracking algorithm to complete the target tracking. We train the Visual Geometry Group (VGG), deep learning model using massive image data, extract the depth feature of tracking targets through the trained VGG deep learning model, and use the depth feature for target detection. The results of the experiment show that, compared with other algorithms such as KCF, the CNNT algorithm achieves a more robust target tracking effect in complex scenes such as illumination change, target rotation, and background clutter.
  Keywords: target tracking; deep learning; convolutional neural network.
  
  
• FCAODNet: a fast freight train image detection model based on embedded FCA  
  by Longxin Zhang, Peng Zhou, Miao Wang, Chengkang Weng, Xiaojun Deng 
  Abstract: The fault detection of freight train image has some problems, such as low detection accuracy and slow detection speed. Aiming at the problem of slow detection speed in the process of train image fault detection, a lightweight object detection model fast channel attention network (FCAODNet) is proposed in this study. FCAODNet consists of four modules, including feature extraction network (FEN), lightweight multiscale feature fusion (LMFF), prediction across scales (PAS), and decoding modules. FEN extracts image features, LMFF fuses features, PAS predicts the location of the target object, and the decoding module obtains the final prediction result. FCAODNet's FEN adopts CSPDarknet53tiny. The designed LMFF is embedded with two FCA modules to improve the detection accuracy. Experiments on train datasets and public datasets show that FCAODNet outperforms other state-of-the-art models in detection speed and has good detection accuracy and robustness.
  Keywords: attention mechanism; fault detection; freight train; object detection.
  
  

Special Issue on: ICNC-FSKD 2021 Cutting-edge High-Performance Computing and Artificial Intelligence Technologies for Medical E-Diagnosis

• Novel freight train image fault detection and classification models based on CNN  
  by Longxin Zhang, Yang Hu, Tianyu Chen, Hong Wen, Peng Zhou, Wenliang Zeng 
  Abstract: Freight train detection systems (TFDS) is deployed to detect the status of freight train components in railway stations, but TFDS is only used to collect, transmit, and store images of train components. Aiming at the problem of train image fault detection of typical cut-out cock, top rod, locking plate, and angle cock of freight trains, a multiclass freight train (MFT) fault recognition model is proposed in this study. First, an object detection model is designed to reduce the dependence on colour and texture, and a bounding box regression method is used to select candidate boxes. Second, a fault classification model is proposed to classify the image segmented by object detection and screen out the image identified as the same fault with the object detection model. The test image dataset of our experiment comes from the China Railway Guangzhou Group Co., Ltd. Experimental results show that the recognition mean accuracy rate (mAR) of the MFT model for typical faults can reach 92.55%, which is 9.83% higher than that of the traditional machine learning method and 5.37% and 3.83% higher than those of faster region-based convolutional neural network (R-CNN) and mask R-CNN, respectively, and has good anti-interference ability for image rotation and noise. In addition, the mAR of MFT on the Modified National Institute of Standards and Technology public dataset can reach 94.60%, and it also has good recognition performance.
  Keywords: convolutional neural network; deep learning; fault detection; freight train fault; image classification.