G.5. Information Technology and Systems Applications
Naga Subrahmanyeswari Nimmakayala; Krishna Prasad M H M
Abstract
Breast cancer detection is critical for early diagnosis and treatment. This paper utilized the BreakHis dataset, comprising 7,907 histopathological images of breast tumors (benign and malignant) captured at varying magnification levels. Initially, a basic CNN was applied, followed by advanced deep learning ...
Read More
Breast cancer detection is critical for early diagnosis and treatment. This paper utilized the BreakHis dataset, comprising 7,907 histopathological images of breast tumors (benign and malignant) captured at varying magnification levels. Initially, a basic CNN was applied, followed by advanced deep learning architectures including ResNet, EfficientNet, Mobilenet, Densenet and VGG19. Among these models, ResNet achieved the highest accuracy of 90.2%. For improving performance, a hybrid combination of hand-crafted features (pHash, HOG, GLCM, Hu Moments, SIFT, ORB and LBP) and transfer learning features (EfficientNet, DenseNet, ResNet, VGG19 MobileNet) was considered. Combined, these features were merged into a single feature vector, and were classified using ML algorithms: Logistic Regression, Naive Bayes, KNN, Decision Tree, Random Forest, Gradient Boosting and XGBoost. XGBoost yielded the highest accuracy of 96.2%. Additionally, deep learning models including Multilayer Perceptron (MLP) and Artificial Neural Networks (ANN) were explored, with ANN slightly outperforming MLP, achieving an accuracy of 98.3% compared to 97.5% for MLP. The results highlight the efficacy of combining traditional and deep learning-based features for improved diagnostic accuracy
H.3. Artificial Intelligence
Mohammad Heydari
Abstract
Deep learning–based super-resolution has become an important tool for enhancing brain magnetic resonance imaging (MRI), particularly when acquisition constraints limit spatial resolution. Lightweight autoencoder architectures have recently been proposed to achieve computational efficiency while ...
Read More
Deep learning–based super-resolution has become an important tool for enhancing brain magnetic resonance imaging (MRI), particularly when acquisition constraints limit spatial resolution. Lightweight autoencoder architectures have recently been proposed to achieve computational efficiency while maintaining reconstruction quality. However, certain architectural choices adopted in these models—most notably the incorporation of encoder–decoder skip connections—raise methodological concerns regarding the preservation of the information bottleneck principle that defines autoencoder-based learning. This note critically examines whether such designs genuinely rely on latent representation learning or instead introduce shortcut pathways that weaken the inferential nature of super-resolution reconstruction.
H.5. Image Processing and Computer Vision
Hamed Modanloujouybari; Yasser Baleghi
Abstract
Accurate estimation of velocity fields from Particle Image Velocimetry (PIV) data is essential for fluid-flow analysis and modeling. PIV relies on Image Processing techniques such as cross-correlation and optical flow to estimate the magnitude and direction of fluid motion; however, traditional approaches ...
Read More
Accurate estimation of velocity fields from Particle Image Velocimetry (PIV) data is essential for fluid-flow analysis and modeling. PIV relies on Image Processing techniques such as cross-correlation and optical flow to estimate the magnitude and direction of fluid motion; however, traditional approaches often struggle in the presence of noise, sharp velocity gradients, and complex flow structures. Recent deep-learning-based methods have shown promising performance, but purely data-driven models may generate physically inconsistent predictions because they do not explicitly incorporate the governing laws of fluid dynamics. In this study, a physics-informed framework based on Recurrent All-Pairs Field Transforms (RAFT), a state-of-the-art deep neural network for dense optical flow estimation, is proposed for velocity-field reconstruction from PIV image sequences. The proposed approach introduces a novel loss function that combines a Charbonnier data term with divergence-free, vorticity-based, and edge-aware smoothness constraints to improve the physical consistency and robustness of the estimated flow fields. The method was evaluated on five benchmark PIV datasets and consistently outperformed the original RAFT model. The average endpoint error (AEE) was reduced from 16.07 to 14.19, corresponding to an improvement of approximately 11.7%, with the largest gains observed in turbulent and complex flow conditions. These results demonstrate that incorporating fluid-dynamics knowledge into deep-learning-based optical flow estimation improves the accuracy, stability, and generalization capability of PIV velocity-field reconstruction.
H.5. Image Processing and Computer Vision
Zahrasadat Sajjadi; Soheil Hamzebeigi; Mohsen Soryani
Abstract
As the global population ages, reliable methods for assessing brain health and age-related changes are increasingly important. Brain age is a promising biomarker of brain health, and machine-learning methods have enabled its estimation from neuroimaging data. However, effective training strategies are ...
Read More
As the global population ages, reliable methods for assessing brain health and age-related changes are increasingly important. Brain age is a promising biomarker of brain health, and machine-learning methods have enabled its estimation from neuroimaging data. However, effective training strategies are required for accurate brain age estimation. This study proposes a two-dimensional convolutional neural network (2D CNN)-based multi-task framework for estimating brain age from magnetic resonance imaging (MRI) scans in the ADNI dataset. The framework uses VGG-16 and U-Net encoder backbones with separate heads for age regression and image-rotation classification. We evaluated the effects of several pretraining strategies, including self-supervised pretraining using the DINO framework and supervised pretraining through brain-tumor segmentation. The best-performing configuration, consisting of a DINO-pretrained VGG-16 backbone and a multi-task prediction head, achieved a mean absolute error (MAE) of 3.27 years, which is competitive with previously reported methods. The results indicate that combining transfer learning with multi-task learning improved performance relative to the corresponding single-task models, suggesting that this combination supports the learning of richer and more generalizable feature representations.
H.6.2.4. Neural nets
Ali Asadi; Morteza Noferesti
Abstract
Hydrogen combustion has emerged as a pivotal technology for decarbonizing the energy sector, offering a clean and sustainable alternative to fossil fuels. This study investigates hydrogen combustion dynamics in a perfectly stirred reactor (PSR) under steady-state, non-premixed conditions. It is employing ...
Read More
Hydrogen combustion has emerged as a pivotal technology for decarbonizing the energy sector, offering a clean and sustainable alternative to fossil fuels. This study investigates hydrogen combustion dynamics in a perfectly stirred reactor (PSR) under steady-state, non-premixed conditions. It is employing CHEMKIN-based simulations to analyze the effects of nitrogen dilution, operating pressure, and equivalence ratio on flame temperature and NOx emissions. The parametric results reveal that nitrogen dilution reduces flame temperature by up to 28% and suppresses NOx emissions by 15–40%, while elevated pressure promotes higher flame temperatures and increased NOx formation. Peak temperature and NOx concentrations are observed under stoichiometric conditions (φ = 1.0), with both quantities decreasing under lean and rich mixture conditions. To enable rapid and accurate prediction of these combustion characteristics, three machine learning models were developed and benchmarked on the CHEMKIN-generated dataset: Gaussian Process Regression, Multilayer Perceptron, and Deep Neural Network. GPR demonstrated the best overall predictive performance, achieving the lowest MAE for temperature prediction (MAE = 1.77) and major species concentrations. Although the DNN produced competitive results (MAE = 58.28), it demanded approximately five times more computational resources than GPR, without a proportional gain in predictive accuracy. All three models, maintained prediction errors below 20% across the investigated parameter space, confirming their viability as efficient tools for hydrogen combustion optimization. These findings demonstrate that physics-informed machine learning models, when combined with high-fidelity combustion simulations, offer a powerful and computationally efficient pathway toward accelerating the design and optimization of clean hydrogen energy systems.
H.3.2.2. Computer vision
Mohammad Hossein Khosravi
Abstract
Document Image Quality Assessment (DIQA) is critical for ensuring the reliability of downstream applications such as Optical Character Recognition (OCR), digital archiving, and automated document workflows. In this paper, we propose a deep learning-based DIQA framework using a Siamese neural network ...
Read More
Document Image Quality Assessment (DIQA) is critical for ensuring the reliability of downstream applications such as Optical Character Recognition (OCR), digital archiving, and automated document workflows. In this paper, we propose a deep learning-based DIQA framework using a Siamese neural network architecture with an InceptionV3 backbone. Our model leverages a composite loss function that combines linear regression loss with a monotonic ranking constraint to jointly optimize for score-level accuracy and perceptual consistency. Unlike prior works that rely on handcrafted features or narrow degradation types, our approach generalizes across diverse distortions commonly observed in scanned and photographed documents. Experimental results on the SOC and SmartDoc-QA datasets demonstrate that the proposed model exhibits a strong correlation with OCR accuracy, achieving SROCC values of 0.952 and 0.873, respectively, and outperforming several state-of-the-art DIQA methods.
H.3.12. Distributed Artificial Intelligence
Samira Mavaddati
Abstract
The rapid growth of intelligent surveillance systems has increased the demand for accurate and efficient criminal activity recognition methods capable of operating in real-world environments. Although conventional deep learning and object detection frameworks have demonstrated promising performance, ...
Read More
The rapid growth of intelligent surveillance systems has increased the demand for accurate and efficient criminal activity recognition methods capable of operating in real-world environments. Although conventional deep learning and object detection frameworks have demonstrated promising performance, they often struggle to capture long-range contextual dependencies and complex interactions present in surveillance scenes. To address these limitations, this study proposes a hybrid deep learning framework that combines the real-time detection capability of YOLOv10 with the global contextual modeling power of Vision Transformers (ViT). An attention-guided feature fusion mechanism is introduced to effectively integrate local spatial representations extracted by YOLOv10 with global semantic features generated by the transformer architecture. The proposed framework is evaluated on the UCF-Crime dataset, which consists of fourteen categories of normal and criminal activities, including burglary, robbery, assault, vandalism, shoplifting, and abuse. Surveillance videos are converted into image sequences and analyzed under two experimental scenarios: (I) a standalone YOLOv10 model and (II) the proposed Attention-Guided YOLOv10-ViT framework. Performance is assessed using accuracy, precision, recall, and F1-score metrics. Experimental results show that the standalone YOLOv10 model achieves an overall classification accuracy of 88.07%, outperforming the previously reported YOLOv8 baseline. More importantly, the proposed hybrid framework attains an accuracy of 93.45%, exceeding both YOLOv10 and earlier YOLOv8-ViT architectures. The improvement is particularly evident in challenging scenarios involving occlusion, illumination changes, cluttered backgrounds, and crowded environments. The results demonstrate that integrating YOLOv10, Transformers, and attention-guided feature fusion provides a scalable, robust, and real-time solution for intelligent surveillance and public monitoring applications.
H.6.5.2. Computer vision
Mahdi Davari; Razieh Rastgoo
Abstract
Detecting driver distraction is critically important, as it remains a major contributor to road accidents and traffic-related injuries worldwide. This study introduces a novel hybrid deep learning model that integrates Spatio-Temporal Graph Convolutional Networks (ST-GCN) with a Transformer Encoder and ...
Read More
Detecting driver distraction is critically important, as it remains a major contributor to road accidents and traffic-related injuries worldwide. This study introduces a novel hybrid deep learning model that integrates Spatio-Temporal Graph Convolutional Networks (ST-GCN) with a Transformer Encoder and Attention mechanisms to effectively detect distracted driving behaviors. The ST-GCN component captures spatial and temporal dependencies in 3D skeletal motion data, modeling the dynamic body movements of the driver. Following this, a Transformer Encoder is employed to further refine temporal representations by leveraging global attention, allowing the model to understand long-range dependencies and subtle behavioral patterns over time. In addition, an Attention mechanism is applied to emphasize the most informative joints and time frames. To address class imbalance in the dataset, the model uses a focal loss function, which helps focus training on more difficult-to-classify examples. The proposed approach is validated on the 3D skeletal Drive&Act dataset, where it achieves a high accuracy of 97.47%, outperforming existing models, particularly under challenging conditions such as poor lighting and complex driving environments. The system demonstrates strong potential for real-time driver monitoring, offering an intelligent solution to enhance road safety and reduce accident risks through early detection of driver distraction.
H.3.2.2. Computer vision
Mohammad Jadidi; Kourosh Kiani; Razieh Rastgoo
Abstract
In recent years, the application of deep learning techniques has revolutionized various domains, including the realm of sports analytics. The analysis of ball tracking and trajectory in sports has become an increasingly vital area of research, driven by advancements in technology and the growing demand ...
Read More
In recent years, the application of deep learning techniques has revolutionized various domains, including the realm of sports analytics. The analysis of ball tracking and trajectory in sports has become an increasingly vital area of research, driven by advancements in technology and the growing demand for data-driven insights in athletic performance. In volleyball, a sport characterized by rapid movements and strategic play, the ability to accurately track the trajectory of the ball is crucial for both training and competitive analysis. This paper proposes novel deep learning models for accurate volleyball ball detection and tracking. By incorporating attention mechanisms into the YOLOv8 and YOLOv10 architecture, our models significantly improve performance, particularly in challenging situations involving occlusions and fast movements. The proposed models across several metrics compared to baseline and other models. Specifically, achieved precision (94.2% and 94.7%, respectively) and recall (88.1% and 87.6%, respectively) and real-time processing speeds, making them suitable for various sports analytics applications.
H.3.2.2. Computer vision
Fatemeh Asadi-Zeydabadi; Ali Afkari-Fahandari; Elham Shabaninia; Hossein Nezamabadi-pour
Abstract
Farsi optical character recognition remains challenging due to the script’s cursive structure, positional glyph variations, and frequent diacritics. This study conducts a comparative evaluation of five foundational deep learning architectures widely used in OCR—two lightweight CRNN based ...
Read More
Farsi optical character recognition remains challenging due to the script’s cursive structure, positional glyph variations, and frequent diacritics. This study conducts a comparative evaluation of five foundational deep learning architectures widely used in OCR—two lightweight CRNN based models aimed at efficient deployment and three Transformer based models designed for advanced contextual modeling—to examine their suitability for the distinct characteristics of Farsi script. Performance was benchmarked on four publicly available datasets: Shotor and IDPL PFOD2 for printed text, and Iranshahr and Sadri for handwritten text, using word level accuracy, parameter count, and computational cost as evaluation criteria. CRNN based models achieved high accuracy on word level datasets—99.42% (Shotor), 97.08% (Iranshahr), 98.86% (Sadri)—while maintaining smaller model sizes and lower computational demands. However, their accuracy dropped to 78.49% on the larger and more diverse line level IDPL PFOD2 dataset. Transformer based models substantially narrowed this performance gap, exhibiting greater robustness to variations in font, style, and layout, with the best model reaching 92.81% on IDPL PFOD2. To the best of our knowledge, this work is among the first comprehensive comparative studies of lightweight CRNN and Transformer based architectures for Farsi OCR, encompassing both printed and handwritten scripts, and establishes a solid performance baseline for future research and deployment strategies.
H.3.2.2. Computer vision
Homayoun Rastegar; Hassan Khotanlou
Abstract
One of the challenges in digital image processing that we face today is the presence of haze in images. This challenge is particularly prominent in imaging areas with humid and rainy weather compared to other locations. Examples of AI-based systems that can be impacted by this type of challenge include ...
Read More
One of the challenges in digital image processing that we face today is the presence of haze in images. This challenge is particularly prominent in imaging areas with humid and rainy weather compared to other locations. Examples of AI-based systems that can be impacted by this type of challenge include smart traffic control cameras, autonomous vehicles, and Video Assistant Referee (VAR) systems in football stadiums, security and surveillance cameras, and more. Therefore, this paper aims to propose a method that can mitigate this problem using Self-Supervised Learning (SSL) and deep learning. To this end, a Convolutional Autoencoder Network (CAN) with Convolutional Block Attention Module (CBAM) was proposed to reduce haze from images. The advantage of the proposed method is using fewer layers and filters compared to other models introduced by previous researchers in this field and using more important convolutional channels and important image regions using CBAM. Experiments in this paper reveal that overusing large or numerous convolutional filters to generate diverse features can reduce a model's ability to dehaze images effectively. Thus, the number of filters should be carefully limited. On the other hand, a combined loss function was used to train the proposed architecture. The proposed model was trained and tested using NH-haze dataset and the Realistic Single Image Dehazing (RESIDE) dataset. To evaluate our method, we used structural similarity index measure (SSIM) and peak signal-to-noise ratio (PSNR). The test results of the proposed architecture showed that it has higher performance compared to the state-of-the-art in the field.
H.3. Artificial Intelligence
Elahe Moradi
Abstract
Liver disorders are among the most common diseases worldwide, and their timely diagnosis and prediction can significantly improve treatment outcomes. In recent years, the application of artificial intelligence, particularly machine learning and deep learning algorithms, in the medical field has gained ...
Read More
Liver disorders are among the most common diseases worldwide, and their timely diagnosis and prediction can significantly improve treatment outcomes. In recent years, the application of artificial intelligence, particularly machine learning and deep learning algorithms, in the medical field has gained tremendous importance and has led to reduced healthcare costs. In this study, the ILPD dataset from the UCI Machine Learning Repository, which comprises 583 liver patient records with 11 features, was utilized. In this research, a predictive framework based on Multilayer Perceptron (MLP) is employed for the prediction of liver disorders. To address the class imbalance in the binary classification dataset, the Synthetic Minority Oversampling Technique (SMOTE)–Tomek approach was implemented to improve data balance. Moreover, due to the presence of a substantial number of outlier values, a robust scaling method was applied for their management. Finally, the performance of the proposed method was compared with three well-known machine learning algorithms. To enhance evaluation robustness, a five-fold cross-validation was employed across all classifiers. All simulations were conducted using Python, and the results illustrate that the proposed method achieves superior performance, with an accuracy of 90.90% compared to state-of-the-art approaches.
H.5. Image Processing and Computer Vision
Sekine Asadi Amiri; Zahra Davoudi
Abstract
wildfires are among the most serious environmental and socio-economic threats worldwide, significantly impacting ecosystems and climate patterns. In recent years, deep learning-based methods, particularly Convolutional Neural Networks (CNNs), have played a crucial role in improving wildfire detection ...
Read More
wildfires are among the most serious environmental and socio-economic threats worldwide, significantly impacting ecosystems and climate patterns. In recent years, deep learning-based methods, particularly Convolutional Neural Networks (CNNs), have played a crucial role in improving wildfire detection accuracy. This study presents an enhanced approach for identifying wildfire-affected areas using deep learning models. Specifically, three models—ResNet50, ResNet101, and EfficientNetB0—have been examined. To improve accuracy and reduce model complexity, the Flatten layer in all three architectures has been replaced with a Global Average Pooling (GAP) layer. This modification reduces the number of features and enhances the extraction of meaningful patterns from images. Additionally, a Dense layer with 128 neurons has been added after the GAP layer to enhance the learning and integration of the extracted features. To prevent overfitting, a Dropout layer with a rate of 0.5 has been incorporated. Finally, a Dense layer with 2 neurons serves as the output layer, responsible for the final classification. These optimizations have led to improved model accuracy and enhanced performance in wildfire detection. The dataset used consists of 42,850 satellite images, categorized into wildfire and nowildfire areas. Experimental results indicate that the ResNet101 model achieved the highest accuracy of 99.60%, while ResNet50 and EfficientNetB0 achieved accuracies of 99.35% and 99.10%, respectively. These results highlight the high potential of deep learning-based methods in improving wildfire detection accuracy and their role in environmental crisis management.
H.3. Artificial Intelligence
Naeimeh Mohammad Karimi; Mehdi Rezaeian
Abstract
In the era of massive data, analyzing bioinformatics fields and discovering its functions are very important. The rate of sequence generation using sequence generation techniques is increasing rapidly, and researchers are faced with many unknown functions. One of the essential operations in bioinformatics ...
Read More
In the era of massive data, analyzing bioinformatics fields and discovering its functions are very important. The rate of sequence generation using sequence generation techniques is increasing rapidly, and researchers are faced with many unknown functions. One of the essential operations in bioinformatics is the classification of sequences to discover unknown proteins. There are two methods to classify sequences: the traditional method and the modern method. The conventional methods use sequence alignment, which has a high computational cost. In the contemporary method, feature extraction is used to classify proteins. In this regard, methods such as DeepFam have been presented. This research is an improvement of the DeepFam model, and the special focus is on extracting the appropriate features to differentiate the sequences of different categories. As the model improved, the features tended to be more generic. The grad-CAM method has been used to analyze the extracted features and interpret improved network layers. Then, we used the fitting vector from the transformer model to check the performance of Grad-CAM. The COG database, a massive database of protein sequences, was used to check the accuracy of the presented method. We have shown that by extracting more efficient features, the conserved regions in the sequences can be discovered more accurately, which helps to classify the proteins better. One of the critical advantages of the presented method is that by increasing the number of categories, the necessary flexibility is maintained, and the classification accuracy in three tests is higher than that of other methods.
H.3. Artificial Intelligence
Rasoul Hosseinzadeh; Mahdi Sadeghzadeh
Abstract
The attention mechanisms have significantly advanced the field of machine learning and deep learning across various domains, including natural language processing, computer vision, and multimodal systems. This paper presents a comprehensive survey of attention mechanisms in Transformer architectures, ...
Read More
The attention mechanisms have significantly advanced the field of machine learning and deep learning across various domains, including natural language processing, computer vision, and multimodal systems. This paper presents a comprehensive survey of attention mechanisms in Transformer architectures, emphasizing their evolution, design variants, and domain-specific applications in NLP, computer vision, and multimodal learning. We categorize attention types by their goals like efficiency, scalability, and interpretability, and provide a comparative analysis of their strengths, limitations, and suitable use cases. This survey also addresses the lack of visual intuitions, offering a clearer taxonomy and discussion of hybrid approaches, such as sparse-hierarchical combinations. In addition to foundational mechanisms, we highlight hybrid approaches, theoretical underpinnings, and practical trade-offs. The paper identifies current challenges in computation, robustness, and transparency, offering a structured classification and proposing future directions. By comparing state-of-the-art techniques, this survey aims to guide researchers in selecting and designing attention mechanisms best suited for specific AI applications, ultimately fostering the development of more efficient, interpretable, and adaptable Transformer-based models.
H.3. Artificial Intelligence
Thomas Njoroge Kinyanjui; Kelvin Mugoye; Rachael Kibuku
Abstract
This paper presents a Multi-Head Self-Attention Fusion Network (MHSA-FN) for real-time crop disease classification, addressing key limitations in existing models, including suboptimal feature extraction, inefficient feature recalibration, and weak multi-scale fusion. Unlike prior works that rely solely ...
Read More
This paper presents a Multi-Head Self-Attention Fusion Network (MHSA-FN) for real-time crop disease classification, addressing key limitations in existing models, including suboptimal feature extraction, inefficient feature recalibration, and weak multi-scale fusion. Unlike prior works that rely solely on CNNs or transformers, MHSA-FN integrates MobileNetV2, EfficientNetV2, and Vision Transformers (ViTs) with a structured multi-level attention framework for enhanced feature learning. A gated fusion mechanism and a Multiscale Fusion Module (MSFM) optimize local texture details and global spatial relationships. The model was trained on a combined dataset of PlantVillage and locally collected images, improving adaptability to real-world conditions. It achieved 98.66% training accuracy and 99.0% test accuracy across 76 disease classes, with 99.34% precision, 99.01% recall, and 99.04% F1 score. McNemar’s test (p = 0.125) and Bayesian superiority probability (0.851) validated its robustness. Confidence variance analysis (0.000010) outperformed existing models, demonstrating MHSA-FN as a scalable, high-performance AI solution for precision agriculture in resource-constrained environments.
H.6.5.13. Signal processing
Khadijeh Aghajani
Abstract
Voice Activity Detection (VAD) plays a vital role in various audio processing applications, such as speech recognition, speech enhancement, telecommunications, satellite phone, and noise reduction. The performance of these systems can be enhanced by utilizing an accurate VAD method. In this paper, multiresolution ...
Read More
Voice Activity Detection (VAD) plays a vital role in various audio processing applications, such as speech recognition, speech enhancement, telecommunications, satellite phone, and noise reduction. The performance of these systems can be enhanced by utilizing an accurate VAD method. In this paper, multiresolution Mel- Frequency Cepstral Coefficients (MRMFCCs), their first and secondorder derivatives (delta and delta2), are extracted from speech signal and fed into a deep model. The proposed model begins with convolutional layers, which are effective in capturing local features and patterns in the data. The captured features are fed into two consecutive multi-head self-attention layers. With the help of these two layers, the model can selectively focus on the most relevant features across the entire input sequence, thus reducing the influence of irrelevant noise. The combination of convolutional layers and self-attention enables the model to capture both local and global context within the speech signal. The model concludes with a dense layer for classification. To evaluate the proposed model, 15 different noise types from the NoiseX-92 corpus have been used to validate the proposed method in noisy condition. The experimental results show that the proposed framework achieves superior performance compared to traditional VAD techniques, even in noisy environments.
H.3. Artificial Intelligence
Pouria Rabiei; Nosratali Ashrafi-Payaman
Abstract
Today, the amount of data with graph structure has increased dramatically. Detecting structural anomalies in the graph, such as nodes and edges whose behavior deviates from the expected behavior of the network, is important in real-world applications. Thus, in our research work, we extract the structural ...
Read More
Today, the amount of data with graph structure has increased dramatically. Detecting structural anomalies in the graph, such as nodes and edges whose behavior deviates from the expected behavior of the network, is important in real-world applications. Thus, in our research work, we extract the structural characteristics of the dynamic graph by using graph convolutional neural networks, then by using temporal neural network Like GRU, we extract the short-term temporalcharacteristics of the dynamic graph and by using the attention mechanism integrated with GRU, long-term temporal dependencies are considered. Finally, by using the neural network classifier, the abnormal edge is detected in each timestamp. Conducted experiments on the two datasets, UC Irvine messages and Digg with three baselines, including Goutlier, Netwalk and CMSketch illustrate our model outperform existing methods in a dynamic graph by 10 and 15% onaverage on the UCI and Digg datasets respectively. We also measured the model with AUC and confusion matrix for 1, 5, and 10 percent anomaly injection.
Document and Text Processing
Zobeir Raisi; Vali Mohammad Nazarzehi
Abstract
The Persian language presents unique challenges for scene text recognition due to its distinctive script. Despite advancements in AI, recognition in non-Latin scripts like Persian still faces difficulties. In this paper, we extend the vanilla transformer architecture to recognize arbitrary shapes of ...
Read More
The Persian language presents unique challenges for scene text recognition due to its distinctive script. Despite advancements in AI, recognition in non-Latin scripts like Persian still faces difficulties. In this paper, we extend the vanilla transformer architecture to recognize arbitrary shapes of Persian text instances. We apply Contextual Position Encoding (CPE) to the baseline transformer architecture to improve the recognition of Persian scripts in wild images, especially for oriented and spaced characters. The CPE utilizes position information to generate contrastive data pairs that help better in capturing Persian characters written in a different direction. Moreover, we evaluate several state-of-the-art deep-learning models using our prepared challenging Persian scene text recognition dataset and develop a transformer-based architecture to enhance recognition accuracy. Our proposed scene text recognition architecture achieves superior word recognition accuracy compared to existing methods on a real-world Persian text dataset.
H.3.2.2. Computer vision
Mobina Talebian; Kourosh Kiani; Razieh Rastgoo
Abstract
Fingerprint verification has emerged as a cornerstone of personal identity authentication. This research introduces a deep learning-based framework for enhancing the accuracy of this critical process. By integrating a pre-trained Inception model with a custom-designed architecture, we propose a model ...
Read More
Fingerprint verification has emerged as a cornerstone of personal identity authentication. This research introduces a deep learning-based framework for enhancing the accuracy of this critical process. By integrating a pre-trained Inception model with a custom-designed architecture, we propose a model that effectively extracts discriminative features from fingerprint images. To this end, the input fingerprint image is aligned to a base fingerprint through minutiae vector comparison. The aligned input fingerprint is then subtracted from the base fingerprint to generate a residual image. This residual image, along with the aligned input fingerprint and the base fingerprint, constitutes the three input channels for a pre-trained Inception model. Our main contribution lies in the alignment of fingerprint minutiae, followed by the construction of a color fingerprint representation. Moreover, we collected a dataset, including 200 fingerprint images corresponding to 20 persons, for fingerprint verification. The proposed method is evaluated on two distinct datasets, demonstrating its superiority over existing state-of-the-art techniques. With a verification accuracy of 99.40% on the public Hong Kong Dataset, our approach establishes a new benchmark in fingerprint verification. This research holds the potential for applications in various domains, including law enforcement, border control, and secure access systems.
H.3.2.2. Computer vision
Zobeir Raisi; Valimohammad Nazarzehi; Rasoul Damani; Esmaeil Sarani
Abstract
This paper explores the performance of various object detection techniques for autonomous vehicle perception by analyzing classical machine learning and recent deep learning models. We evaluate three classical methods, including PCA, HOG, and HOG alongside different versions of the SVM classifier, and ...
Read More
This paper explores the performance of various object detection techniques for autonomous vehicle perception by analyzing classical machine learning and recent deep learning models. We evaluate three classical methods, including PCA, HOG, and HOG alongside different versions of the SVM classifier, and five deep-learning models, including Faster-RCNN, SSD, YOLOv3, YOLOv5, and YOLOv9 models using the benchmark INRIA dataset. The experimental results show that although classical methods such as HOG + Gaussian SVM outperform other classical approaches, they are outperformed by deep learning techniques. Furthermore, Classical methods have limitations in detecting partially occluded, distant objects and complex clothing challenges, while recent deep-learning models are more efficient and provide better performance (YOLOv9) on these challenges.
H.5. Image Processing and Computer Vision
Sekine Asadi Amiri; Fatemeh Mohammady
Abstract
Fungal infections, capable of establishing in various tissues and organs, are responsible for many human diseases that can lead to serious complications. The initial step in diagnosing fungal infections typically involves the examination of microscopic images. Direct microscopic examination using potassium ...
Read More
Fungal infections, capable of establishing in various tissues and organs, are responsible for many human diseases that can lead to serious complications. The initial step in diagnosing fungal infections typically involves the examination of microscopic images. Direct microscopic examination using potassium hydroxide is commonly employed as a screening method for diagnosing superficial fungal infections. Although this type of examination is quicker than other diagnostic methods, the evaluation of a complete sample can be time-consuming. Moreover, the diagnostic accuracy of these methods may vary depending on the skill of the practitioner and does not guarantee full reliability. This paper introduces a novel approach for diagnosing fungal infections using a modified VGG19 deep learning architecture. The method incorporates two significant changes: replacing the Flatten layer with Global Average Pooling (GAP) to reduce feature count and model complexity, thereby enhancing the extraction of significant features from images. Additionally, a Dense layer with 1024 neurons is added post-GAP, enabling the model to better learn and integrate these features. The Defungi microscopic dataset was used for training and evaluating the model. The proposed method can identify fungal diseases with an accuracy of 97%, significantly outperforming the best existing method, which achieved an accuracy of 92.49%. This method not only significantly outperforms existing methods, but also, given its high accuracy, is valuable in the field of diagnosing fungal infections. This work demonstrates that the use of deep learning in diagnosing fungal diseases can lead to a substantial improvement in the quality of health services.
H.6.2. Models
Simon Kawuma; Elias Kumbakumba; Vicent Mabirizi; Deborah Nanjebe; Kenneth Mworozi; Adolf Oyesigye Mukama; Lydia Kyasimire
Abstract
Tuberculosis (TB) is an underestimated cause of death in children, with only 45% of cases correctly diagnosed and reported. It is estimated that 1.12 million TB cases occurred among newborns, children, and adolescents aged less or equal 14 years. In Uganda, TB prevalence is 8.5% in children and 16.7% ...
Read More
Tuberculosis (TB) is an underestimated cause of death in children, with only 45% of cases correctly diagnosed and reported. It is estimated that 1.12 million TB cases occurred among newborns, children, and adolescents aged less or equal 14 years. In Uganda, TB prevalence is 8.5% in children and 16.7% in adolescents. Treatment and diagnosing TB is difficulty and its high mortality rate is due to many gaps in the diagnosis of this illness especially among children. As a strategy to curb TB mortality rate in children, there exist a need to improve and expedite the screening for TB among children. Chest X-ray (CXR) are commonly used in TB burden countries like Uganda to diagnose TB patients but interpretation of the patients’ radiograph needs skilled radiologists who are few. To this end, this research aims to close the TB mortality gap in children by applying AI, primarily deep learning techniques, to detect TB in children. The study created five models, one from scratch and four transfer learning and were trained and verified using digital CXR radiograph images of children who visit the TB clinic at Mbarara Regional Referral Hospital. The model classifies clinical images of patients into normal or Tuberculosis. Transfer learning models; VGG16, VGG19, Inception V3, and ResNet50 outperformed scratch model with validation accuracy of 79.91%, 69.21%, 53.0%, 51.09% and 50.01% respectively. We hope that once the deep learning models are implemented and adopted by the radiologist, it will reduce the time spent by radiologist while analyzing CXR images.
H.5. Image Processing and Computer Vision
Farima Fakouri; Mohsen Nikpour; Abbas Soleymani Amiri
Abstract
Due to the increased mortality caused by brain tumors, accurate and fast diagnosis of brain tumors is necessary to implement the treatment of this disease. In this research, brain tumor classification performed using a network based on ResNet architecture in MRI images. MRI images that available in the ...
Read More
Due to the increased mortality caused by brain tumors, accurate and fast diagnosis of brain tumors is necessary to implement the treatment of this disease. In this research, brain tumor classification performed using a network based on ResNet architecture in MRI images. MRI images that available in the cancer image archive database included 159 patients. First, two filters called median and Gaussian filters were used to improve the quality of the images. An edge detection operator is also used to identify the edges of the image. Second, the proposed network was first trained with the original images of the database, then with Gaussian filtered and Median filtered images. Finally, accuracy, specificity and sensitivity criteria have been used to evaluate the results. Proposed method in this study was lead to 87.21%, 90.35% and 93.86% accuracy for original, Gaussian filtered and Median filtered images. Also, the sensitivity and specificity was calculated 82.3% and 84.3% for the original images, respectively. Sensitivity for Gaussian and Median filtered images was calculated 90.8% and 91.57%, respectively and specificity was calculated 93.01% and 93.36%, respectively. As a conclusion, image processing approaches in preprocessing stage should be investigated to improve the performance of deep learning networks.
H.5. Image Processing and Computer Vision
Mohammad Mahdi Nakhaie; Sasan Karamizadeh; Mohammad Ebrahim Shiri; Kambiz Badie
Abstract
Lung cancer is a highly serious illness, and detecting cancer cells early significantly enhances patients' chances of recovery. Doctors regularly examine a large number of CT scan images, which can lead to fatigue and errors. Therefore, there is a need to create a tool that can automatically detect and ...
Read More
Lung cancer is a highly serious illness, and detecting cancer cells early significantly enhances patients' chances of recovery. Doctors regularly examine a large number of CT scan images, which can lead to fatigue and errors. Therefore, there is a need to create a tool that can automatically detect and classify lung nodules in their early stages. Computer-aided diagnosis systems, often employing image processing and machine learning techniques, assist radiologists in identifying and categorizing these nodules. Previous studies have often used complex models or pre-trained networks that demand significant computational power and a long time to execute. Our goal is to achieve accurate diagnosis without the need for extensive computational resources. We introduce a simple convolutional neural network with only two convolution layers, capable of accurately classifying nodules without requiring advanced computing capabilities. We conducted training and validation on two datasets, LIDC-IDRI and LUNA16, achieving impressive accuracies of 99.7% and 97.52%, respectively. These results demonstrate the superior accuracy of our proposed model compared to state-of-the-art research papers.
