International Journal of Innovative Computing and Applications (19 papers in press)

Regular Issues

• Implementation of Cloud Service Platform for Monitoring Charging Facility Status of Electric Vehicle based on MQTT  
  by Lei Li, Weidong Liu, Xiaohui Li, Guang Yang, Dan Li 
  Abstract: Aiming at the heterogeneous charging facilities of many manufacturers, the difficulty of communication and information interaction among these devices, and the lack of real-time information sharing, this paper makes full use of Message Queuing Telemetry Transport(MQTT) which is a lightweight message publishing and subscribing protocol, designs a cloud service platform for monitoring the status of charging facilities of electric vehicles, and collects monitoring data of charging facilities of electric vehicles to store in cloud server. Firstly, the network topology structure of cloud service platform is given. Then, the logical architecture and functions of the platform are designed. The information collection, pushing process and security design of the cloud service platform are described in detail. Finally, the implementation and performance tests of the presented system are carried out. The MQTT-based Internet of Things Middleware in the cloud platform can provide remote parameter configuration and information acquisition functions for charging facilities in smart grid environment, shield the underlying hardware devices, realize data interaction between the sensor layer and the upper application, and realize the integration of multi-source information of heterogeneous devices. The research results have important application value for improving the automation level of electric vehicle charging facilities monitoring and the real-time information interaction among heterogeneous devices of electric vehicles.
  Keywords: Electric Vehicle; Charging Facility Monitoring; Cloud Computing; MQTT; Middleware.
  
  
• New Architectural Optical Character Recognition Approach for Cursive Fonts: The Historical Maghrebian Font as an Example  
  by Ilyes OULED OMAR, Sofiene HABOUBI, Faouzi BENZARTI 
  Abstract: In this paper, we intend to present different built Maghrebian font databases with giving the different challenges facing the development of an Optical Character Recognition system able to treat it.rnAlso, the full architecture able to treat the historical Maghrebian font is revealed. Further, a complete design with the accuracy of each module is provided. The novel OCR architecture includes a binarization module based on deep neural networks with an accuracy of 98.1%. Moreover, it involves three segmentation tasks based on deep learning approaches for text/non-text separation, columns division and connected components segmentations. The classification task is based on the DenseNet model with an accuracy of 98.95%. the post-processing module is based also based on deep learning approaches based on sequential modelling with an accuracy of 81.3%. It also includes a user-feedback stage with an accuracy of 94.7%. The total system accuracy is 89.06%.
  Keywords: OCR; Cursive Historical Documents; Maghrebian Font Database; Deep Learning.
  
  
• New delay-independent exponential stability rule of delayed Cohen-Grossberg neural networks  
  by Cheng-De Zheng, Haorui Meng, Shengzhou Liu 
  Abstract: This manuscript studies the stability for a class of Cohen-Grossberg neural networks (CGNNs) with variable delays. By practicing the scheme of Lyapunov function (LF), M-matrix (MM) theory, homeomorphism theory and nonlinear measure (NM) method, a new sufficient condition is obtained to ensure the existence, uniqueness and global exponential stability (GES) of equilibrium point (EP) for the studied network. As the condition independent of the delay, it can be applied to networks with large delays. The result generalizes and improves the earlier publications. Finally, an example is supplied to exhibit the power of the results and less conservativeness over some earlier publications.
  Keywords: stability; inequality; delay; homeomorphism.
  
  
• Fuzzy improved Firefly based MapReduce for Association Rule Mining  
  by Lydia Nahla Driff, Habiba Drias 
  Abstract: In a research environment, Biology-derived algorithms founded on neighboring local optimum are considered among the most powerful optimization algorithms. The biggest challenge of these algorithms is to be able to ensure balanced convergence and this is what we describe in this paper, we were interested in devising an improved version of FireFly Algorithm (FF) for Association Rules Mining called IFF. In order to refine generated rules based on frequent patterns, we had to reduce or even eliminate blind mating from the design of genetic algorithm and replaced it by mating between mature fireflies. Knowing that it could not have been possible to refine ARM using classic method such as Apriori, proposed approach has more advanced methods such as controlled genetic operations to manipulate frequent patterns, and the uses of fuzzy logic to control IFF parameters and assure convergence calibration based on data size, algorithm iterations and temporary local optimum. On the other hand, we executed IFF under Hadoop to get a MapReduce system and ensure the most optimal execution time. To analyze the quality of our proposal, we made simulations on MEDLINE collection using statistical analysis and comparing with classical algorithms and recent evolutionary approaches. Results indicate that the proposed approach is superior to existing algorithms with an accuracy of 10% to 50% and save execution time around 36% while ensuring a good balance between the quality and variety of the obtained knowledge.
  Keywords: Swarm Intelligence; Firefly algorithm; Genetic algorithm; Fuzzy logic; Association Rules Mining; Frequent patterns; MapReduce; Hadoop; MEDLINE collection.
  
  
• Fuzzy Modeling Techniques for Improving Multi-label Classification of Software Bugs  
  by Rama Ranjan Panda, Naresh Kumar Nagwani 
  Abstract: Software bug repositories stores a wealth of information related to the problems that occurred during the software development. Today's software development is a modular approach, with multiple developers working in different locations all around the world. A software bug may belong to multiple categories and can be resolved by more than one developer. For understanding the multiple causes of software bugs and proper bug information management at large bug repositories, better classification of software bugs is needed. In the proposed work, a multi-label fuzzy system-based classification (ML-FBC) is proposed. A fuzzy system is used to compute the membership of software bugs into multiple categories. Then a fuzzy C-means clustering algorithm is used to create various clusters. Once the clusters are created, the cluster-category mapping is done for various software bugs. For a new bug, the fuzzy similarity values are computed, and the created cluster-category mappings are utilized to categorize it. Using a user-defined threshold value, a new bug is classified into multi-label categories. Experiments are carried out on available benchmark data sets to compare the performance measures F1 score, BEP score, Hloss, accuracy, training time, and testing time of various multi-label classifiers. The proposed ML-FBC outperforms existing multi-label classifiers.
  Keywords: Mining Bug Repositories; Bug Information Management; Fuzzy Modeling; Multi-Class Categorization; Multi-Label Classification.
  
  
• HARDeep: Design and Evaluation of a Deep Ensemble Model for Human Activity Recognition  
  by R. Raja Subramanian, V. Vasudevan 
  Abstract: With the emergence of smartness in various fields including medical science, forensics and security, remote monitoring of human activities has gained more interests in research. The ambulatory health monitoring services includes monitoring the activities of mentally challenged and elderly people. These activities are tracked to get rid of abnormalities. In this research paper, we propose a novel framework for activity recognition from video sequences captured from static cameras and those captured from UAVs. The proposed framework, HARDeep, composes of two models: a human detection model leveraging YOLOv3, to extract the set of video frames containing human and an activity recognition model leveraging the ensemble of three deep learning models: GoogleNet, ResNet-50 and ResNet-101. HARDeep is evaluated against three datasets including Hollywood2 dataset, KTH dataset and the UCF-ARG dataset, consisting of video sequences captured from UAVs. Empirical evaluations state that HARDeep exhibits a sound recognition accuracy compared to similar inference models.
  Keywords: Activity recognition; Deep ensemble model; UAV; Scene Stabilization; Fog Computing.
  DOI: 10.1504/IJICA.2023.10052593
   
  
• Automatic Speech Recognition of Gujarati Digits using Wavelet Coefficients in Machine Learning algorithms  
  by Purnima Pandit, Shardav Bhatt 
  Abstract: In today's world, Automatic Speech Recognition (ASR) is an important task implemented via Machine Learning (ML) to assist Artificial Intelligence (AI). It has diverse applications such as humanmachine interactions, handsfree computing, voice search, domestic appliance control and many more. Speech recognition in an Indian regional language becomes a very necessary task in order to facilitate people, who can communicate only using their mother tongue and the disabled ones. In this article, we have proposed and performed experiments of speech recognition for Gujarati language, particularly for Gujarati digits. The recorded speech is preprocessed and then speech features are extracted from it using Mel-Frequency Discrete Wavelet Coefficient (MFDWC). These features are trained using Artificial Neural Networks (ANN) for classification. Two ANN architectures namely, Multi-layer Perceptrons (MLP) and Radial Basis Function Networks (RBFN) are used for training and recognition. The experimental results obtained in this work are compared with our previous experimental results.
  Keywords: Automatic Speech recognition (ASR); Machine Learning (ML); Artificial Neural Networks (ANN); Radial Basis Function Networks (RBFN).
  
  
• A New Dispersive Flies Optimisation Algorithm for the Sum of Three Cubes  
  by Boian Lazov, Tsvetan Vetsov 
  Abstract: By first solving the equation $x^3+y^3+z^3=k$ with fixed $k$ for $z$ and then considering the distance to the nearest integer function of the result, we turn the sum of three cubes problem into an optimisation one. To our knowledge, this is a novel approach. We then present a modification of the dispersive flies optimisation (DFO) algorithm and apply it to this function in the case with $k=2$. We have two goals: to show the viability of using optimisation when searching for integer solutions and to measure how efficient our modified DFO is. We have significantly improved the performance of DFO for very large and discrete search spaces by adding new mechanisms to increase the exploration behaviour of the flies. As a comparison we also use two implementations of simulated annealing. The efficiency of the algorithms is measured by their running times. We model the data by assuming two underlying probability distributions -- exponential and log-normal, and calculate relevant numerical characteristics for them. Finally, we evaluate the statistical distinguishability of our methods with respect to some standard parametric and non-parametric statistical tests.
  Keywords: dispersive flies optimisation; particle swarm optimisation; Diophantine equations; sum of three cubes; simulated annealing.
  
  
• Prioritizing Test Cases to Improve the Software Fault Detection using MCDM Methods  
  by Maryam Mohammadi Sarpiri, Keyvan Mohebbi, Ali Jamshidi 
  Abstract: To decrease the cost of software testing, we can run a subset of test cases, but this may result in residual faults. To keep the efficiency of testing, the most important test cases should be selected through a prioritization approach. Such prioritization requires the assessment of different criteria, so it can be formulated as a multi-criteria decision-making (MCDM) problem. This research proposes an approach to select the proper subset of test cases using the MCDM methods. Three MCDM methods, namely, Fuzzy SAW, Fuzzy VIKOR, and Fuzzy TOPSIS are applied to prioritize the test cases concerning various criteria. To select a subset of test cases, a threshold is determined for different pairs of the most important criteria. The proposed approach is applied to an actual e-government software system with two variants. The experimental evaluations indicate the efficiency of this approach with respect to both the failure rate and the average percentage of fault detection metrics.
  Keywords: Software Testing; Test Case Prioritization; Multi-Criteria Decision Making; Fault Failure Rate; and Average Percentage Fault Detection.
  
  
• Deep Learning Intelligence for Influencer-based Topological Classification for Online Social Networks  
  by Somya Jain, Adwitiya Sinha 
  Abstract: Social network analysis provides quantifiable methods and topological metrics to examine the networked structure for several interdisciplinary applications. In our research, a social network of GitHub community is constructed that forms a dense network of 37700 developers with 289003 associations amongst them. The research involves finding the central developers in the GitHub network using graph analytics and benchmark centrality metrics, including Degree, Betweenness, Closeness, PageRank and Eigenvector; which is based upon network structural information. Our research methodology provides a breakthrough towards predicting the classification of GitHub users using artificial intelligence-based learning model trained with derived topological network centrality metrics. The proposed approach performs feature extraction for the developers by computing centrality score of each user followed by building correlation matrix using centrality parameters based on network topology. Further, the derived topological centrality scores were used as input features to train and build artificial intelligence-based models for classification. Our experimentation shows better performance of artificial neural network over autoencoders, logistic regression and hyper-parameter tuned support vector machine. Certain intermediate outcomes include correlation, principal component analysis, loss monitoring, etc. The performance evaluation was performed in terms of macro and weighted F1-score, recall, precision, and accuracy.
  Keywords: Online social network; GitHub community; user influence computing; network centrality; artificial intelligence; topological classification.
  
  
• Fuzzy based MPPT Control for solar PV Applications  
  by RAVINDRANATH TAGORE YADLAPALLI  
  Abstract: Nowadays, the solar energy is playing a vital role as compared to the fossil fuels due to their several advantages such as more reliable, inexhaustible, low running cost, creating job opportunities and energy independence. However, the solar based power generation is affected by the atmospheric conditions. Therefore, the imperative aspect is to track maximum power from the solar PV panel with respect to the changing environmental conditions. This paper mainly focuses on the various MPPT techniques for the solar panel. It includes the implementation of the MPPT algorithms based on the Perturb & Observe (P&O), incremental-conductance (INC) and fuzzy logic control (FLC). At first, the solar PV panel is modelled based on the single-diode model. The P-V and I-V curves are shown with different solar irradiation variations. The main objective here is to analyse the dynamics in terms of the tracking time as well as the steady state error. The exhaustive simulation results are presented for the boost converter based on the INC, P&O and FLC MPPT algorithms.
  Keywords: solar panel; dynamics; fuzzy control.
  DOI: 10.1504/IJICA.2023.10051410
   
  

Special Issue on: IWSSIP 2020 Latest Scientific and Theoretical Advances in Systems, Signals and Image Processing

• Performance Evaluation of Energy Reconstruction Methods for the ATLAS Hadronic Calorimeter Using Collision Data  
  by Juan Marin 
  Abstract: Particle accelerators are machines that collide particle beams next to the speed of light. The Large Hadron Collider, the largest and most powerful beam collider, operates at a center-of-mass energy of 13~TeV with a 40~MHz collision rate. In the ATLAS experiment, optimal filter techniques can be applied on energy estimation even in the occurrence of noise and signal overlapping. In this context, the present work compares the performance of the energy estimation in the main hadronic ATLAS calorimeter, for two methods: the baseline algorithm OF2 (Optimal Filter) and COF (Constrained Optimal Filter). The work proposes an adaptive estimator of the energy signal pedestal in the ATLAS experiment at the LHC. For performance comparisons, the statistics from the energy estimation distributions are employed. The results show that the COF method presents a better performance than the OF2 method in terms of estimation error. The proposed signal baseline estimator properly improves the COF efficiency.
  Keywords: High-energy physics; Signal estimation; Optimal filtering; Signal pile-up.
  
  
• Light Leaf Spots Segmentation Algorithm based on Color Difference Vectors  
  by Sheng Miao, Weili Kou, Lihui Zhong 
  Abstract: Leaf spots image segmentation and quantifying is a crucial step of disease image recognition, which is very important for monitoring plant diseases and subsequently diagnosis. In the leaf spots segmentation, compared with dark spots, light leaf spots segmentation is hinder due to unobvious color characteristics, and easily confused with leaf veins. This study specially focuses on the light leaf spots segmentation and vein removal. First of all, a mask extraction model was constructed to get leaf spots mask, includes three steps: L * a * b color model mapping, color difference vectors construction and color difference vectors segmentation. Secondly, veins mask and spots mask have been extracted from healthy leaves and spots leaves by mask extraction model respectively. And vein ratio of vein mask was calculated to evaluate whether vein removal is needed or not. Thirdly, vein area and roundness were calculated for both vein mask and spots mask, and remove areas with high similarity to vein features from spots mask. Leaves with light spots of 12 plants types were selected to test proposed algorithms, the experimental results show: compared to the manual segmentation method for identifying light leaf spots, the presented algorithm has higher average accuracy (86%) than both traditional watershed (77%), Barbedo 2017(82%) and CNN (81%). Moreover, this algorithm is easy to be used in other plant species.
  Keywords: Light leaf spots; Image segmentation; Vein removal; Color difference vector.
  
  
• Detecting Acute Leukemia in Blood Slides Images Using a CNNs Ensemble  
  by Maíla Claro, Rodrigo Veras, Luis Vogado, André Santana, Vinicius Machado, Justino Santos, João Tavares 
  Abstract: Leukemia is a disease that has no defined etiology and affects the production of white blood cells in the bone marrow. Young cells or blasts are produced abnormally, replacing normal blood cells (white, red blood cells, and platelets). Consequently, the person suffers problems in transporting oxygen and infections combat. Acute leukemia is a particular type of leukemia that causes abnormal cell growth in a short period, requiring a quick start of treatment. Classifying the types of acute leukemia in blood slide images is a vital process, and a system of assisting doctors in selecting treatment becomes necessary. This article presents an ensemble approach composed of three convolutional neural networks (CNNs) - Alert Net-RWD, Resnet50 and InceptionV3. These CNNs, individually, demonstrated effectiveness in differentiating blood slides with Acute Lymphoid Leukemia (ALL), Acute Myeloid Leukemia (AML), and Healthy Blood Slides (HBS). We verified that the union of these three well-known CNNs improves the hit rates of current techniques from the literature. The experiments were carried out using 18 public data sets with 3,293 images, and the proposed CNNs ensemble achieved an accuracy of 96.17%, and precision of 96.38%
  Keywords: Acute leukemia diagnosis; model ensemble; convolutional neural network.
  
  
• Evaluation of Banknote Identification Methodologies Based on Local and Deep Features  
  by Leonardo P. Sousa, Rodrigo M. S. Veras, Luis H. S. Vogado, Laurindo S. Britto Neto, Romuere R. V. Silva, Flávio H. D. Araujo, Fátima N. S. Medeiros 
  Abstract: There are many people with disabilities; it is estimated that 39 million people are blind and 246 million have limited vision, giving a total of 285 million visually impaired people. Information and communication technologies can help disabled people achieve greater independence, quality of life, and inclusion in social activities by increasing, maintaining, or improving their functional capacities. This paper presents a significant evaluation of local and deep features for an automatic methodology for identifying banknotes. To determine the best local features, we evaluated a set of four point-of-interest detectors, two descriptors, seven ways of generating the image signature, and six classification methodologies. To define the deep features, we extract features using three pre-trained well-known CNNs. Additionally, we evaluated using a hybrid descriptor formed by the combination of local and deep features. In this situation, the features were selected according to their gain ratios and used as input to the classifier. Experiments performed on US Dollar (USD), Euro (EUR), and Brazilian Real Banknotes (BRL) obtained rates of accuracy of 99.96%, 99.12%, and 96.92%, respectively.
  Keywords: accessibility; visually impaired; banknote recognition; assistive technologies.
  
  
• Multilevel CNN for Anterior Chamber Angle Classification using AS-OCT Images  
  by Marcos Ferreira, Geraldo Braz, João Almeida, Anselmo Paiva, Rodrigo Veras 
  Abstract: Glaucoma is the second leading cause of blindness and the first leading cause of irreversible blindness. The main types of the disease are open-angle and angle-closure glaucoma. In people with angle-closure, the anterior chamber angle is narrow, which leads to a rising intraocular pressure, and consequently, optic nerve damage, causing vision loss. Since it is irreversible, an early diagnosis is essential. So, angle classification is fundamental for diagnosis. Anterior segment optical coherence tomography is one of the imaging tests used to diagnose the disease. In addition to the fact that no eye contact is necessary, this test provides a fast way to analyze the anterior chamber angle, to classify it as open or closed. We propose the anterior chamber angle classification method based on visual feature extraction, using deep neural networks in this work. In a multilevel architecture, different pre-trained CNNs are adjusted to extract deep features and train two classifiers. The best model extracted visual features from the anterior camera angle in the experiments, and achieved a sensitivity value of 1.000 as the best result.
  Keywords: Angle-Closure Glaucoma; Transfer Learning; Deep Features; Multilevel Networks.
  
  
• Transfer Learning Based Lung Segmentation and Pneumonia Detection for Pediatric Chest X-Ray Images  
  by Vandecia Fernandes, Gabriel Bras, Lisle Faray De Paiva, Geraldo Braz Junior, Anselmo Cardoso De Paiva, Luis Rivero 
  Abstract: Pneumonia is the leading cause of morbidity and mortality in under-five children, especially in developing countries. Accordingly to UNICEF and World Health Organization, a child dies of pneumonia every 39 seconds, and pneumonia kills more children than any other infectious disease, accounting for 15% of all deaths of children under five years old. In regions with a high prevalence, the early detection and treatment of pneumonia can significantly reduce children's mortality rates. Commonly, a chest x-ray is a diagnostic exam. Nevertheless, it is a problematic image for reading and interprets, requiring an expert physician. So, it is essential to provide computational methods to help exam interpretation or enhance important information. This paper proposes a transfer learning method to segment lung regions on the Chest X-Ray dataset to extract ROI for pneumonia detection. The results are promising and reach 0.917 of dice using U-Net combined with InceptionV3 in a chest x-ray dataset without lung annotation. For pneumonia detection, the method achieves 0.954 precision.
  Keywords: Pneumonia; Transfer Learning; Deep Learning; Segmentation.
  
  
• Wind Turbine Fault Detection: A Semi-Supervised Learning Approach With Two Different Dimensionality Reduction Techniques  
  by Fernando Sá, Rafaelli Coutinho, Eduardo Ogasawara, Diego Brandao, Rodrigo Toso 
  Abstract: The quest to save the environment has led many countries to change their mix of energy sources, with most countries focusing on wind energy as a clean, alternative source. Since wind turbines are at the center of this revolution, ensuring their continuous operation by preemptively detecting and correcting faults is a key to success. Towards that end, making use live operational signals from the turbine, various data-driven methods for fault prediction using traditional machine learning have been proposed. As with all traditional machine learning solutions, these require careful feature selection and hyperparameter tuning for each turbine, which is not simple to scale. This work adopts a novel, automatic, end-to-end AutoML approach covering aspects from features and hyperparameters selection to fault prediction using semi-supervised support vector machines guided by the multi-objective optimization framework Non-dominated Sorting Genetic Algorithm II (NSGA-II). Experiments were carried out using a dataset containing unlabelled records of five 2.0MW wind turbines. In the results, we analyze and compare the approaches with respect to fault detection performance. We found that our AutoML approach using NSGA-II for feature selection offers up to 9% improvement in solution quality over the state-of-the-art while being fully automated and requiring no costly and time-consuming feature engineering.
  Keywords: Feature Selection; Fault Detection; Wind Turbine; Semi-Supervised Learning.
  
  
• Heuristic-based approaches for fracture detection in borehole images  
  by Maira Moran, Eduardo Vasconcellos, Jordan Cuno, Mauro Biondi, Jose Riveaux, Maury Correia, Esteban Clua, Aura Conci 
  Abstract: The success of an oil well drilling process depends, among other factors, on the borehole stability, which is closely related to fractures. Borehole imaging tools are commonly used to identify this kind of feature. Visually, fractures can be identified as sinusoidal curves in the 2D data obtained by the Ultrasonic Borehole Imager (UBI). Their automatic detection is not trivial. Previous works in literature proposed solutions for this problem. Most of them are based on extensive curve searches with a high computational cost. A recent work uses a more efficient method based on a heuristic algorithm. However, it results in many false fractures. In this work, we propose a set of 11 new heuristics-based innovative methods, composed of two different algorithms and three pre-processings, to identify the sinusoidal curves related to fractures in UBI data. Moreover, we include a post-processing step to reduce the number of incorrect fractures. We perform an evaluation of these proposed methods in real data, comparing them to the only heuristic-based method previously presented in the literature. Our achieved results are very promising, with the best methods presenting fewer false curves and a higher rate of correctly detected fractures.
  Keywords: Oil drilling; Ultrasonic Borehole Image; Fractures; Sinusoidal curves; Heuristics; Variable Neighborhood Descendent; Iterated Local Search; Detection.