International Journal of Computer Applications in Technology (30 papers in press)

Regular Issues

• FPGA implementation and Multisim simulation of a new four-dimensional two-scroll hyperchaotic system with coexisting attractors  
  by Sundarapandian Vaidyanathan, Esteban Tlelo-Cuautle, Khaled Benkouider, Aceng Sambas, Ciro Fabian Bermudez-Marquez, Samy Abdelwahab Safaan 
  Abstract: Field-programmable gate array (FPGA) design of a new four-dimensional two-scroll hyperchaotic system is investigated in this work. A detailed system modelling of the new system with a hyperchaotic attractor begins this work with phase plots, which is followed by a bifurcation study of the new system. Special dynamic properties such as multistability and symmetry are also investigated for the new system. Using Multisim software, a circuit model is designed and simulated for the new hyperchaotic system. FPGA design and Multisim simulation of the new system enable practical applications in science and engineering. The implementation of the FPGA design in this work is carried out by applying two numerical schemes, viz. Forward Euler and Trapezoidal methods. Experimental attractors observed in the oscilloscope show good match with the Matlab signal plots.The FPGA hardware resources are detailed for both numerical methods.
  Keywords: hyperchaos; bifurcation; symmetry; phase plots; hyperchaotic system;rnparameters; stability; multistability; circuit model; FPGA implementation.
  
  
• Improving hybrid-layer convolutional neural network system for lung cancer nodule classification using enhanced weight optimisation algorithm  
  by Vikul Pawar, P. Premchand 
  Abstract: In recent times, lung cancer is evolving as a highly life-threatening disease for human beings. According to the WHO, lung cancer disease is the second largest cause of deaths as compared to all other types of cancer. The prevailing available technology is striving to get more exposure in the field of medical science using Computer Assisted Diagnosis (CAD), where image processing is playing a crucial role for detecting the cancerous nodules in computer tomographic images. Augmenting the machine learning techniques with image processing algorithms is becoming a more comprehensive examination of cancer disease in proposed CAD systems. This paper is describes a heuristic approach for lung cancer nodule detection, and the proposed model predominantly consists of the following tasks, which are image enhancement, segmenting ROI (Region of Interest), features extraction, and nodule classification. In pre-processing, primarily the Adaptive Median Filter (AMF) filtering method is applied to eliminate the speckle noise from input CT images of Lung Image Database Consortium (LIDC) and Image Database Resource Initiative (IDRI): in the LIDC-IDRI dataset, the quality of input image is improved by applying Histogram Equalization (HE) technique with Contrast-Limited Adaptive (CLA) approach. Secondly, in the successive stage the Improved Level-Set (ILS) algorithm is used to segment the ROI. Furthermore, the third step of the projected work is applied to extract the definite learnable texture features and statistical features from the segmented ROI. The extracted features in the subsequent stage of classification are applied to Hybrid-Layer Convolutional Neural Network (HL-CNN) architecture to classify the lung cancer nodule as either benign or malignant. Principally this research is carried out by contributing to each stage of it, where the novel concept of the improved Hybrid-Layer Convolutional Neural Network (HL-CNN) is employed by optimising and selecting the optimal weight using the Enhanced Cat Swarm Optimisation (ECSO) algorithm. The experimental result of the proposed HL-CNN using the weight optimisation algorithm ECSO is achieved an accuracy of 93%, which is comparatively efficient with respect to existing models such as DBN, SVM, CNN, WOA, MFO, and CSO. Moreover, the proposed model conclusively gives a decision on the detected nodule as either benign or malignant.
  Keywords: Computer Assisted Diagnosis (CAD); Computer Vision; Cancer Diagnosis; Image Classification; Image Enhancement; Image Segmentation; Feature Extraction.
  
  
• Prediction model for total amount of coke oven gas generation based on FCM-RBF  
  by Lili Feng, Jun Peng, Zhaojun Huang 
  Abstract: The rational use of Coke Oven Gas (COG) is of great significance to improve the economic efficiency of enterprises. In this paper, a COG generation prediction model based on fuzzy C-mean clustering (FCM) and radial basis function (RBF) neural network is proposed to address the problems such as the difficulty of accurate modelling of COG generation process and the difficulty of real-time flow prediction. Firstly, the coke oven production process is analysed and correlation analysis is used to select the influencing factors. Secondly, the FCM is used to classify the working conditions of the coke oven, and the appropriate number of working conditions is selected through experiments. Finally, the prediction models under different working conditions are established separately by using RBF. The experiments were carried out using actual industrial production data, and the experimental results showed that the model could provide guidance reference for the dispatchers.
  Keywords: coking oven process; fuzzy C-means clustering; prediction model; radial basis function neural network.
  
  
• Advancing financial security: integrating AI and blockchain for cloud network protection in supply chain financing  
  by Weishuang Xu, Daming Li 
  Abstract: This paper investigates the integration of Artificial Intelligence (AI) into cloud network security protocols, recognizing its capacity to analyze vast datasets and detect patterns effectively. By leveraging AI, cloud network security can be fortified, thereby enhancing risk identification and control within enterprise supply chain financing, a domain where efficient data collection and integration remain pivotal challenges. The study explores the application of blockchain technology (BT) in enterprise supply chain systems, demonstrating its potential to enable real-time monitoring and management of the financing process. The research findings indicate substantial improvements in risk identification (0.99) and security performance (1.22), affirming the feasibility of integrating IoT-based BT into enterprise supply chain financing for enhanced risk management and security.
  Keywords: corporate supply chain; risk identification and control; internet of things; blockchain; electricity system trading.
  DOI: 10.1504/IJCAT.2025.10069456
   
  
• Development of intelligent data collection and management system based on internet of things big data crawler technology  
  by Jin Chen, Yao Li, Xia Hua, Long Lu 
  Abstract: This paper used web crawler technology to develop intelligent data collection and management system. This paper first analyzed the basic principles of system design and the structural requirements of the system according to the system requirements, and then evaluated the overall results of the intelligent data collection and management system. The process and principle of information collection using the IoT Big Data (BD) crawler technology were introduced in detail. Finally, the information collection effect of the system was verified by experiments. The experimental results showed that the data acquisition accuracy of the system was high, accounting for more than 90%. The system had high usability and efficiency; users were satisfied with the system, and data collection and management could be carried out well.
  Keywords: intelligent data collection; data management; big data of the internet of things; big data crawler technology.
  DOI: 10.1504/IJCAT.2025.10069544
   
  
• Visual evaluation of tourism ecological and environmental protection green management in the context of sustainable development  
  by Kun Zheng, Lijing Zhang 
  Abstract: The tourism industry has developed well in the past few years. However, due to the further spread of the COVID-19 in recent years, the development of tourism industry in various regions is relatively poor. At the same time, some behaviors of tourists in tourism have also brought a greater burden to the ecological environment of various regions. This is mainly due to the differences of the ecological environment between different regions, which leads to some habitual or other uncivilized behaviors that would cause a destructive blow to the local ecological environment. However, with the increasing scale of the tourism industry, people’s economic level and material conditions are getting better and better, and more and more people want to visit the cultural scenery of different regions, which leads to the increasingly obvious impact of the tourism industry on the ecological environment of different regions.
  Keywords: tourism ecology; green management; sustainable development; visualisation analysis.
  DOI: 10.1504/IJCAT.2025.10069655
   
  
• A lightweight authentication architecture for local communication of power terminal based on quantum key  
  by Chao Chen, Jin Qian, Shaojie Luo 
  Abstract: In recent years, there has been a notable acceleration in the construction of new power systems, particularly those incorporating renewable energy sources. At present, the scheme to bolster the secure communication capabilities of power terminals through a fusion of quantum cryptography and lightweight identification key technology is grappling with practical deployment challenges. To address this predicament, this study introduces a novel lightweight authentication architecture for power terminals within local communication networks, anchored on quantum key principles. By leveraging a secure encryption chip in conjunction with lightweight authentication key technology, the proposed architecture facilitates efficient and secure connectivity among a multitude of terminals in local communication settings. A comprehensive security analysis of the proposed solution is also presented.
  Keywords: local communication; quantum cryptography; secure chip; lightweight authentication key technology; authentication.
  DOI: 10.1504/IJCAT.2025.10070796
   
  
• Construction of intelligent tourism public service platform featuring communication big data and internet of things  
  by Shuo Liang, Yiran Wang, Lili Liu 
  Abstract: Tourism has evolved into a modern service industry and strategic pillar, driven by advancements in information and communication technology. This transformation has ushered in the era of intelligent tourism. The study introduces the background of intelligent tourism, summarising academic research on tourism public service platforms, big data (BD), and the Internet of Things (IoT). An algorithm model is proposed to establish a theoretical framework for constructing an intelligent tourism platform characterised by BD and IoT. Factor analysis identifies key elements for platform development, followed by simulation experiments to validate its effectiveness. Results demonstrate that the proposed platform increases user satisfaction by 20% compared to traditional systems. In the BD era, with growing socioeconomic and technological progress, integrating information technology across industries has become critical. Developing intelligent tourism platforms represents a significant theoretical and practical challenge for the tourism industrys advancement.
  Keywords: intelligent tourism; big data; internet of things; communication technology; wireless communication.
  DOI: 10.1504/IJCAT.2025.10071195
   
  
• Transforming design elements and enhancing visual communication via machine learning-based graphic design  
  by Zhenyuan Hao, Xiaoyang Shen 
  Abstract: This paper proposes convolutional efficient transformer-based image feature extraction. First, it combines convolution from different angles to introduce translation invariance, scale invariance, and locality. Then, a lightweight convolution module with depthwise convolution and atrous convolution is used to change the traditional processing of the input image of the transformer to accelerate the convergence speed and improve the stability. Aiming at the problem that cycle generative adversarial network (CycleGAN) has low texture clarity when processing images, the convolutional efficient transformer-based image feature extraction algorithm is added to the generator of GAN to enhance the effect of CycleGAN in image style transfer. Results demonstrate that the proposed algorithm performs well regarding top-1 accuracy, number of parameters, floating-point operations, and convergence rate. It can effectively analyze and classify visual elements, providing valuable insights and aiding in tasks such as image categorization, style transfer, generative design, or automated design recommendation systems.
  Keywords: machine learning; graphic design; transformer; CycleGAN; Image style transfer.
  DOI: 10.1504/IJCAT.2025.10071201
   
  
• RNN-PSO: a tuned neural network with optimisation algorithm for keratoconus classification  
  by G.P. Ramesh, Priyanka Subramanian, B. Ramakantha Reddy 
  Abstract: Keratoconus is the medical illness wherein the cornea attains a conical shape caused by thinning of corneal stroma. Symptoms based on keratoconus stage gets unnoticed. The advanced phases are found through vision loss and protrusion. The diagnosis of Keratoconus and its severity classification from corneal topography images has gained importance with advanced imaging and machine learning. Convolutional Neural Networks (CNN) is used for extracting significant features from images categorize the keratoconus. Then, extracted features are classified using Recurrent Neural Network (RNN) and the hyper parameters of the RNN is optimized using Particle Swarm Optimization (PSO) which improves the keratoconus classification performance. The performance of proposed method is evaluated with four classes namely normal, subclinical, keratoconus and advanced keratoconus in terms of accuracy, sensitivity and specificity. The method attained the accuracy of 0.80 on normal, 0.81 on subclinical, 0.85 on keratoconus and 0.91 on advanced keratoconus.
  Keywords: convolutional neural networks; keratoconus; particle swarm optimisation recurrent neural network; sub-clinical.
  DOI: 10.1504/IJCAT.2024.10071229
   
  
• Location selection of intelligent warehouse combined with swarm intelligence  
  by Jinghui Wu, Junhua Yan 
  Abstract: The location selection of an intelligent warehouse is a critical decision that affects its efficiency and profitability. This paper introduces a new method that combines swarm intelligence with geographic information systems to tackle this challenge. The approach models the issue as a multi-objective optimization problem and employs a swarm intelligence algorithm to determine the best locations for setting up a warehouse. The algorithm considers several factors, including the proximity to customers, the availability of transportation infrastructure, and the cost of land and labor. To assess the efficacy of the proposed method, the algorithm was implemented in a real-world scenario involving a logistics company as a case study. The results demonstrate that the proposed approach can effectively identify optimal warehouse locations that improve the efficiency and profitability of the logistics operations.
  Keywords: location selection; intelligent warehouse; swarm intelligence; geographic information systems.
  DOI: 10.1504/IJCAT.2025.10071337
   
  
• MOOC system platform based on edge computing and artificial intelligence  
  by Bifeng Li, Lilibeth Cuison 
  Abstract: Online users flexibly obtain learning resources on MOOC system platform. Nowadays learners need to spend more time to screen the relevant content of the curriculum. Based on this, it aims to use artificial intelligence technology to assist online education. A large amount of video data brings high computing load and low real-time performance to the cloud server. Therefore, it is feasible to combine edge computing and artificial intelligence by taking advantage of the characteristics that the edge end has certain computing power and low latency near the terminal. This paper proposes a hybrid recommendation model Deep-AM based on artificial intelligence, and deploys it to the edge server. Thus, it can perform real-time detection and feature extraction. Compared with traditional models LFM, NARRE and DeepCoNN, it is shown that the proposed Deep-AM has faster response speed and higher task completion rate. And it has better practicability when applied to MOOC system.
  Keywords: edge computing; artificial intelligence; MOOC system; deep-AM model.
  DOI: 10.1504/IJCAT.2025.10071338
   
  
• A secure and reliable large scale online physical education solution based on deep learning for cloud computing  
  by Wenbo Song 
  Abstract: With the continuous expansion of online education scale, its existing problems have gradually emerged. Therefore, it is necessary to develop a secure and reliable large scale online physical education solution, cultivate higher-order thinking and guide students to learn effectively. This article studies cloud computing and deep learning models. And it proposes an improved model based on graph knowledge tracing(GKT). The improved model can better discover causal relationships and use network structure to track students' knowledge status. Finally, experiments are conducted on the improved model and the classical model, proving that the improved model based on GKT performs well. The improved model based on GKT outperforms other models and has superior convergence speed and accuracy. It has good practicality for large scale online physical education, helps to obtain interactive information of implicit knowledge in the scheme, and increases the richness of education.
  Keywords: online physical education; cloud computing; deep learning; GKT.
  DOI: 10.1504/IJCAT.2025.10071339
   
  
• Electrochemical energy storage power stations decision-making via digital twins and simulation-based data fusion  
  by Zhoubo Weng, Yimin Deng, Zhiyong Zhao, Shanshan Zhao 
  Abstract: The digital twin model for power station uses a dynamic three-dimensional representation to map the physical system and real-time data, encompassing monitoring control, state evolution, analysis, evaluation, and integrating real-time operational laws, power station analysis, and inference logic. This enables real-time monitoring, operational management, intelligent analysis, virtual inspection, and simulation training. Moreover, the joint Kalman filter is employed for data fusion to enhance the fusion effect of heterogeneous data from multiple sources within the digital twin-based electrochemical energy storage power station. Simulation results demonstrate promising performance in terms of fusion error and efficiency. By leveraging accurate data fusion, the proposed data-driven digital twin for electrochemical energy storage power stations offers several benefits, including improved accuracy, operational efficiency, proactive maintenance, real-time monitoring, enhanced system reliability, and safety. These advantages significantly contribute to optimizing the data fusion process in electrochemical energy storage power stations, ultimately leading to enhanced performance and decision-making.
  Keywords: electrochemical energy storage power stations; digital twins; data-driven decision; data fusion; joint Kalman filter.
  DOI: 10.1504/IJCAT.2025.10071340
   
  
• DRN-MCOA: image deblurring using deep residual network with modified coot optimisation algorithm  
  by Godekere Shivashankar Yogananda, Ananda Babu Jayachandra, Ahmed Alkhayyat, Dayananda Pruthviraja 
  Abstract: In this manuscript, a hybrid model is introduced for effective image deblurring. A Deep Residual Network (DRN) is implemented for reducing artificial traces that results in pleasant denoised images. Secondly, a Modified Coot Optimization Algorithm (MCOA) is incorporated with the DRN for selecting optimal kernel and threshold parameters. The exploitation and exploration ability of the MCOA is improved by employing an opposition based learning method and Cauchy mutation This process resolves the problem of local optima and improves the convergence rate. This DRN-MCOA model’s efficacy is investigated on real time images and RealBlur dataset. The DRN-MCOA model obtained a Peak Signal to Noise Ratio (PSNR) of 33.40 dB and a Structural Similarity Index (SSIM) of 0.96 on a real time collected image. Correspondingly, it achieved PSNR of 30.34 dB and 37.55 dB and SSIM of 0.92 and 0.96 on the RealBlur-J and RealBlur-R dataset.
  Keywords: coot optimisation algorithm; deep residual network; image deblurring; image processing; restoration.
  DOI: 10.1504/IJCAT.2025.10071789
   
  
• Research on image encryption algorithm based on Logistic chaotic system  
  by Qing Lu, Te Zhang, Junxiang Wan, Siyuan Xu 
  Abstract: For the recently proposed image encryption algorithm [1] based on the Logistic chaotic system, the algorithm first generates chaotic sequences and plaintext images through the key for XOR operations, and then pixel chaos and block chaos according to the chaotic sequence, and finally obtains the ciphertext image. Through analysis, it is found that the correlation between plaintext and ciphertext images in this algorithm is not high, and there is no ciphertext feedback mechanism. This algorithm can be attacked by selecting ciphertext. In this paper, the algorithm is improved. And the relevant test analysis of the improved algorithm is carried out, and the results are better than the algorithm proposed by Zhang and others.
  Keywords: chaotic system; image encryption; chosen-ciphertext attack.
  DOI: 10.1504/IJCAT.2025.10072127
   
  
• Hie-Graph-YOLOv9: a hierarchical YOLOv9 model with graph-based SE attention mechanism for vehicle detection in complex background
  by T. Selvamuthukumar, K. Vijayalakshmi, P. Dhanalakshmi, R. Abinaya 
  Abstract: Advanced vehicle detection algorithms are key to Intelligent Transportation Systems (ITS), enabling real-time traffic analysis, congestion and security management. Existing models like YOLOv9 face challenges in feature selection and learning, especially in dynamic or cluttered environments. To address these limitations, this research proposes Hie-Graph-YOLOv9 which is an extended version of YOLOv9 based on improving the feature selecting, feature learning and loss function by incorporating Hiera Transformers, Graph-based GAN-SE attention mechanism and Geometric-based Weighted Smooth L1 loss function. Hiera Transformers, integrated into the backbone network across four stages, refine multi-scale feature learning, ensuring robust representation of fine-grained and global patterns. The Graph-based GAN-SE, embedded in the bottleneck module, emphasises critical regions of feature maps, enhancing detection accuracy. Additionally, a Geometric-based Weighted Smooth L1 loss function is employed for bounding box regression, improving convergence speed and training stability. Experimental evaluations demonstrate the superiority of Hie-Graph-YOLOv9, achieving an AP (0.5) of 79.5%, improvement of faster convergence by 120 Epochs and an increased inference speed of 41.95 FPS, outperforming state-of-the-art models. This work offers a significant step forward in vehicle detection under complex real-world conditions.
  Keywords: object detection; YOLO; vehicle; Hiera; graph; squeeze and excitation.
  
  
• Hie-Graph-YOLOv9: A Hierarchical YOLOv9 model with Graph-based SE attention mechanism for vehicle detection in complex background  
  by T. Selvamuthukumar, K. Vijayalakshmi, P. Dhanalakshmi, R. Abinaya 
  Abstract: Advanced vehicle detection algorithms are key to intelligent transportation systems (ITS), enabling real-time traffic analysis, congestion and security management. The proposed Hie-Graph-YOLOv9 method is an extended version of YOLOv9 based on improving the feature selecting, feature learning and loss function. In this YOLO architecture, we induced Hiera Transformers in the backbone network in four stages for improving the feature learning. We also introduced the Graph based GAN-SE attention mechanism in the bottleneck module for giving attention to essential feature map regions and utilized Geometric based Weighted Smooth L1 loss function for bounding box prediction for faster convergence, training stability and improved accuracy.
  Keywords: object detection; YOLO; vehicle;Hiera; graph; Squeeze and Excitation.
  DOI: 10.1504/IJCAT.2025.10072853
   
  
• An unsupervised video summarisation method based on temporal convolutional networks  
  by Ke Jin, Hui Li, Haoran Li, Qichuang Liu, Rong Chen, Shikai Guo 
  Abstract: Video summarisation automatically selects a sparse subset of video frames that best represent the semantic content of the input video. Previous work mainly utilised Long Short-Term Memory (LSTM) networks to learn how to assess the importance of each video frame and then select appropriate frames to compose a video summary. However, these models still have shortcomings, such as limited memory capacity in handling long-term dependencies, and extended training time. Therefore, we present a deep video summarisation model, named TCN-SUM, which centres around Bi-TCN and models the frame sequence. TCN-SUM consists of three modules: frame selection module incorporates both Bidirectional Temporal Convolutional Network (Bi-TCN) and self-attention to model inter-frame dependencies, video reconstruction module reconstructs the original video based on the summary, and discriminator module measures the similarity between the original and reconstructed videos. Experimental studies on two benchmark data sets demonstrate that TCN-SUM outperforms state-of-the-art techniques, achieving superior performance in unsupervised approaches and showing competitiveness compared to supervised methods.
  Keywords: Bi-TCN; video summarisation; self-attention; LSTM.
  DOI: 10.1504/IJCAT.2024.10067536
   
  
• Artificial intelligence-based drug structure extraction and representation in chemistry documents  
  by Xin Xu, Jiaheng Pan, Dazhou Li 
  Abstract: The feature extraction and representation of chemical bond linear molecular structures from chemistry publication images is of great significance to rediscover the properties of chemical structure, but the rule-based method and the existing deep learning methods are facing the problem of low recognition rate. This paper presents ChemRAL, an automatic conversion model for chemical molecular images and identifiers. ChemRAL employs an encoder-decoder architecture with a ResNet residual network for image feature extraction, and an attention-based LSTM long-term memory network for converting molecular structure images into chemical identifiers. Comparative evaluations demonstrate that the ChemRAL model outperforms existing methods in terms of cross entropy loss and accuracy of the longest common subsequence. The conducted experiments have successfully showcased the advancements achieved by the ChemRAL model. The findings unequivocally indicate that ChemRAL not only enhances the precision and effectiveness of molecular image feature extraction and representation but also provides a significant benchmark.
  Keywords: artificial intelligence; drug discovery; drug structure extraction and representation; image characteristics extraction; drug information representation; LSTM.
  DOI: 10.1504/IJCAT.2024.10068749
   
  
• Research on machine reading comprehension for BERT and its variant neural network models  
  by YanFeng Wang, Ning Ma, Wenrong Lv 
  Abstract: Currently, machine reading comprehension models primarily rely on the LSTM networks with the gate mechanism. In the present study, we employ BERT and its variant pre-training-trained language model to conduct research and experimentation on the DuReader data set. We find that enhanced masking techniques, such as full-word mask masking and dynamic mask masking, can significantly enhance the model's performance in machine reading comprehension tasks. Therefore, the ROUGE-L and BLEU-4 values of RoBERTRoBERTa-wwm-ext model on the test set achieve ROUGE-L and BLEU-4 scores of 51.02% and 48.14%, respectively, which are 19.12% and 8.94% higher than the benchmark model. In addition, addressing the issue of suboptimal model performance is not optimal when the data is dealing with large data and the effective information is relatively dispersed. This paper employs a three-step pre-processing approach for the data set. This method is based on the F1-score to identify relevant paragraphs, answer modules and feature precomputation, so that the performance of precompute features, ultimately bringing the pre-training-trained language model, is shown to bring the model's performance closer to the average human reading comprehension level.
  Keywords: machine reading comprehension; BERT pre-training language model; masking mode.
  DOI: 10.1504/IJCAT.2024.10063844
   
  
• Text Swin Transformer: a new transformer model for enterprise management text classification  
  by Bo Yuan, Yang Wang, Jing Chi 
  Abstract: As the Internet and information technology advance at a swift pace, enterprises have amassed vast quantities of data stemming from their operational activities, managerial processes, and other business functions. Behind these data lies extremely valuable knowledge that is closely related to the enterprise, which can provide critical decision-making support for enterprise management. Therefore, how to discover valuable knowledge from voluminous textual data, transform these textual data into a genuine asset for enterprises, and leverage this asset to inform decision-making and propel their growth, has become a challenging task. Inspired by this, this paper revises the Swin Transformer model, which has achieved wide success in computer vision tasks, and proposes a Text Swin Transformer (TST) model for enterprise management text data classification tasks. Specifically, we revise the original window attention, shifted window attention and patch merging strategies to text window attention, text shift window attention and window scaling strategies, in order to meet the needs of text classification tasks. In contrast to the state-of-the-art approaches in text classification task, the proposed model not only achieves excellent results on multiple standard datasets, but also requires less computation and boasts a faster computation speed.
  Keywords: enterprise management text classification; Text Swin Transformer; window attention; shift window attention.
  DOI: 10.1504/IJCAT.2024.10069181
   
  
• Building heating management based on deep learning to strength intelligent building design  
  by Yipeng Wang, Shilin Zhang 
  Abstract: Building heating systems based on deep networks plays an important role in intelligent urban construction, providing a comfortable environment for residents. An accurate heating prediction system which collects time-series signals through various sensors can grasp the temperature of buildings in advance, avoiding unnecessary energy waste. Traditional convolution operations only model the local dependency relationships of time-series signals for temperature prediction. To improve the long-range dependencies of features, long short-term memory networks (LSTM) and Transformer-based models have been widely used in building heating prediction tasks. However, local dependencies cannot be utilised, and these models require a large amount of computation. To this end, this paper proposes a novel variant Swin-Transformer-based model by exploiting a sparse self-attention mechanism for building temperature prediction. In addition, we introduce the traditional convolution operations for local dependency modelling. We conducted extensive experiments on our self-built dataset. All experimental results show that the proposed model achieves higher performance than LSTM and Transformer models in terms of temperature prediction.
  Keywords: building heating system; deep learning; sparse self-attention; LSTM; Transformer.
  DOI: 10.1504/IJCAT.2024.10068871
   
  
• Assisted English oral teaching with mouth recognition technology  
  by Shaoli Xiong, Rui Cong, Lin Siew Eng, Chunyan Ruan 
  Abstract: Deep learning-based speech emotion analysis has been widely applied in English oral teaching. Due to the close relationship between oral pronunciation and mouth shape, mouth recognition has received increasing attention to assist speech recognition. To effectively improve the effectiveness of English oral teaching, this paper constructs an efficient emotion analysis model by introducing human mouth-shape features. Specifically, we first use the Dlib tool to locate 68 facial landmarks and crop mouth-associated landmark information. To improve the extraction of mouth landmark features, we introduce the spatiotemporal graph convolutional network which can effectively mine spatial and temporal features from landmark information. In addition, to effectively model local and global dependencies, we utilise Focal-Transformer for speech feature extraction. To verify the effectiveness of our proposed model, we conducted extensive comparative experiments on two publicly available multimodal sentiment analysis datasets and the self-built English oral teaching dataset. All the experimental results confirm that our proposed model obtains a higher performance compared to other deep models.
  Keywords: deep learning; mouth recognition; English oral teaching; multimodal emotion analysis; facial landmarks; focal-transformer.
  DOI: 10.1504/IJCAT.2024.10068792
   
  
• Deep one-class classification induced authentication for security protection of wearable IoT application  
  by Meng Tian, Guimei Liu, Lijing Xie 
  Abstract: The Internet of Things (IoT) has become one of the most popular directions in the field of network and communication. However, with the rapid development of IoT, the number of terminal devices shows a geometric order of magnitude growth. The sensor network composed of sensing terminal devices can obtain information in all aspects, but at the same time, malicious attackers can also obtain the information. In order to solve this issue, this paper designs a lightweight authentication for security protection in wearable IoT environment through edge-cloud architecture. The proposed lightweight authentication system adopts PPG signal as biometric feature, which can be easily obtained and hard to forge. The authentication is implemented by using deep one-class classification model which is trained by using a PPG signal library. In order to avoid the limitation of resources at edge nodes, the deep one-class classification models are deployed at the cloud server. The experiments show that the proposed authentication system can achieve a promising result.
  Keywords: authentication; IoT security protection; PPG signal; deep one-class classification.
  DOI: 10.1504/IJCAT.2025.10071395
   
  
• Intelligent optimisation induced 5G spectrum dynamic allocation  
  by Yuan Zhou 
  Abstract: The 5G communication network system provides services with large capacity, high speed and low latency, which may make the shortage of spectrum resources become more and more serious. However, most of the existing spectrum allocation algorithms are based on continuous spectrum allocation and do not consider the needs of users. The continuous spectrum allocation will generate many spectrum fragments that cannot satisfy user's requirements. These spectrum fragments cannot be fully utilised, which results in a waste of spectrum. In order to tackle this issue, this paper adopts a spectrum allocation algorithm that can jointly consider both the needs of users and discontinuous spectrum aggregation. In this algorithm, non-contiguous spectrum fragments can be aggregated to meet the user demand of spectrum, which can make full use of small spectrum fragments to avoid waste and improve spectrum utilisation.
  Keywords: 5G spectrum; spectrum allocation; spectrum segment; deep learning; distributed spectrum allocation.
  DOI: 10.1504/IJCAT.2025.10071396
   
  
• Optimal routing strategy for multi-constraint scenarios in software-defined optical transport networks  
  by Peng Zhu, Hong Sun, Qian Xiang, Zhenming Zhang 
  Abstract: To address the challenge of computing the optimal loop-free path under multiple complex logical constraints in a software-defined optical transport network with hybrid optical-electrical layer scenarios, this paper proposes an optimal path calculation method for complex logic combination constraints. First, a unified constraint expression is employed to describe the constraints, simplifying and decomposing the logical relationships. The network topology is then transformed accordingly, mapping various complex constraints onto the original structure. Next, an improved K-shortest path algorithm is applied to obtain, in a single computation, the path that satisfies multiple complex constraints, including 'AND', 'OR' and bidirectional link requirements, while ensuring the global optimal solution. Experimental results demonstrate that this method offers higher practical value compared to other algorithms.
  Keywords: software-defined optical transport network; SDOTN; unified constraint expression; complex constraints; must pass links; loop-free path; path calculation; hierarchical topology; KSP; K-shortest path.
  DOI: 10.1504/IJCAT.2025.10071714
   
  
• Research on multi-scale face detection based on graph embedded Swin-Transformer  
  by Fang Wang, Huang Zhong 
  Abstract: The existing Swin-Transformer has achieved great success in the field of object detection due to its lightweight, hierarchical and efficient long-distance dependency modelling advantages. However, in complex scenes, the impact of scale changes seriously restricts the performance of face detection. Therefore, this article proposes a multi-scale face detection network based on graph embedding Swin-Transformer. This network fully utilises multi-scale features and contextual information, effectively improving the performance of multi-scale face detection. Specifically, we designed a multi-feature fusion module based on Swin Transformer, which effectively integrates deep features of advanced information with shallow texture features. In addition, to effectively model the contextual information of the target, we use embedding graph convolution operations. We conducted extensive comparative experiments on publicly available data sets, and the experimental results showed that the proposed model achieved higher recognition performance.
  Keywords: multi-scale face detection; Swin-Transformer; graph convolution embedded; multi-scale feature fusion.
  DOI: 10.1504/IJCAT.2025.10072633
   
  
• A novel leader replacement-based unmanned aerial vehicle flocking scheme  
  by Junling Shi, Guoyu Zhu, Aihua Men, Guiying Meng 
  Abstract: The flocking motion is a fundamental and crucial operation in multi-UAV systems, encompassing navigation and obstacle avoidance. However, the unknown and random environment poses a great challenge to traditional Unmanned Aerial Vehicle (UAV) flocking and navigation control methods. This paper uses Reinforcement Learning (RL) techniques to achieve navigation and obstacle avoidance for a swarm of UAVs in unknown environments. The RL algorithm employed in this study is MAPPO, which has been integrated with an attention mechanism to form the ATT-MAPPO algorithm. This incorporation of the attention mechanism enables the agent to effectively filter out irrelevant information and focus its attention solely on crucial features relevant to the task at hand, thereby significantly enhancing decision-making accuracy and efficiency. The leader replacement strategy effectively balances the energy distribution within the UAV flocking, thereby significantly extending the overall flight range of the UAV. Finally, we demonstrate the scalability and adaptability of ATT-MAPPO in a simulation experiment.
  Keywords: flocking; multi-agent reinforcement learning; unmanned aerial vehicles; leader replacement.
  DOI: 10.1504/IJCAT.2025.10072643
   
  
• Retrieving similar images: using TriCLR+CC-CDLBP features extraction algorithm  
  by P. John Bosco, S. Janakiraman 
  Abstract: The colour-based operator extracts the visual information of an image by analysing the values of neighbouring pixels. Both single and multiple colour features can be derived from various colour attributes, which can serve as image feature descriptors in certain applications. This paper presents a novel feature representation method, TriCLR + CC-CDLBP, which integrates the primary tricolour (TriCLR) approach in colour spaces such as RGB, YCbCr, and L*a*b* with the Change the Central Pixel and Change the Direction (CC-CDLBP) method in LBP. This approach leverages multiple colour features in different colour spaces to enhance image feature representation. A hybrid algorithm is designed to improve the performance of similar image retrieval by applying the primary Tricolour (TriCLR) method to individual channels R, Y, and L* incorporating the CC-CDLBP method to refine feature extraction. By combining TriCLR + CC-CDLBP, our method introduces a robust hybrid approach that outperforms existing image retrieval techniques.
  Keywords: colour features; RGB; texture features; CBIR; multiple features; local binary pattern.
  DOI: 10.1504/IJCAT.2025.10072655