Mohd. Aquib Ansari
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Mohd. Aquib Ansari has a total of 71 co-authors. Here are the top co-authors by number of shared publications:
- Shahnawaz Ahmad
Bennett University
20 shared publications - Dushyant Kumar Singh
Motilal Nehru National Institute of Technology (MNNIT)
12 shared publications - Arvind Mewada
Bennett University
10 shared publications Dushyant Kumar Singh
MNNIT Allahabad, India
7 shared publicationsArvind Mewada
Bennett University, India
5 shared publications- Lalit Kumar
Department of Computer Science & Engineering, Acharya Narendra Deva University of Agriculture and Technology, Ayodhya, India
5 shared publications
Latest Publications
- Open Access
- 30 September 2026
- Published by Sakarya University Journal of Computer and Information Sciences in Sakarya University Journal of Computer and Information Sciences
- Vol. 9 (4) , 1166-1178
Automated Diagnosis of Tomato Leaf Diseases with Attention-Enhanced Feature Aggregation
Tomato crop yield can be enhanced by using advanced agricultural technologies when plant leaf diseases are detected early. This article proposes a new tomato disease classification model, called Multiscale Parallel Feature Aggregation Network with Attention Fusion (MPFAN-AF), that classifies diseases from leaf images. This model comprises parallel convolutional branches that extract multiscale features to obtain rich data, which are then fused through an attention-based fusion module that performs global average pooling and weights the channels. This model enables the network to optimize patterns associated with the disease while disregarding the diseased area’s background noise. Generalization performance is improved with dropout and L2 weight decay. The refined features are passed through a lightweight multi-layer perceptron for classification. Evaluated on a benchmark tomato leaf disease dataset, MPFAN-AF outperforms conventional Convolutional Neural Networks (CNNs) and existing attention-based models across accuracy, precision, recall, and F1-score. Overall, MPFAN-AF delivers an efficient, accurate, and interpretable solution for automated disease diagnosis in precision agriculture.
- 30 September 2026
- Published by Sakarya University Journal of Computer and Information Sciences in Sakarya University Journal of Computer and Information Sciences
- Vol. 9 (4) , 1166-1178
Open AccessAutomated Diagnosis of Tomato Leaf Diseases with Attention-Enhanced Feature Aggregation
Tomato crop yield can be enhanced by using advanced agricultural technologies when plant leaf diseases are detected early. This article proposes a new tomato disease classification model, called Multiscale Parallel Feature Aggregation Network with Attention Fusion (MPFAN-AF), that classifies diseases from leaf images. This model comprises parallel convolutional branches that extract multiscale features to obtain rich data, which are then fused through an attention-based fusion module that performs global average pooling and weights the channels. This model enables the network to optimize patterns associated with the disease while disregarding the diseased area’s background noise. Generalization performance is improved with dropout and L2 weight decay. The refined features are passed through a lightweight multi-layer perceptron for classification. Evaluated on a benchmark tomato leaf disease dataset, MPFAN-AF outperforms conventional Convolutional Neural Networks (CNNs) and existing attention-based models across accuracy, precision, recall, and F1-score. Overall, MPFAN-AF delivers an efficient, accurate, and interpretable solution for automated disease diagnosis in precision agriculture.
- 16 July 2026
- Published by Wiley
- p. 123-134
Content‐aware Convolutional VAE for Anime Face Synthesis
- 16 July 2026
- Published by Wiley
- p. 89-107
Advanced Foundations and Future Trends in Generative AI for Visual Media
- 16 July 2026
- Published by Wiley
- p. 89-107
Advanced Foundations and Future Trends in Generative AI for Visual Media
- 16 July 2026
- Published by Wiley
- p. 59-72
Raindrop Removal in Images and Videos Using Generative AI : A Survey
- 16 July 2026
- Published by Wiley
- p. 59-72
Raindrop Removal in Images and Videos Using Generative AI : A Survey
- 16 July 2026
- Published by Wiley
- p. 1-22
Introduction to AI ‐generated Image and Video Synthesis
- 1 June 2026
- Published by Pleiades Publishing Ltd in Pattern Recognition and Image Analysis
- Vol. 36 (2) , 272-284
ShopSafe: Artificial Intelligence-Powered Theft Prevention Using Multiscale Analysis
Shoplifting has become a growing problem for businesses, large and small, whether they are standalone stores or major retail stores. It is usually practiced by people hiding items in the clothes or bags they carry away without paying at the premises. However, no automated approach to this problem is available yet. Therefore, we proposed an artificial intelligence monitoring system that autonomously recognizes shoplifting incidents in retail environments by identifying object-stealing actions in video sequences to address this issue. When such events are detected, the system generates a notification. Our approach proposes a multiscale parallel feature aggregation network to extract salient features in different video sequences. Then, a long short-term memory model was combined to categorize activities and detect shoplifting events. This work makes a significant contribution by optimizing the recurrent learning classifier parameters in relation to sequence length and batch size. We also made available a novel shoplifting dataset, created in a lab-style environment and containing synthesized video clips of normal and shoplifting activities. The proposed system obtained 91.48% accuracy, surpassing existing methods and highlighting its potential as a robust and effective solution for shoplifting detection.
- 1 June 2026
- Published by Pleiades Publishing Ltd in Pattern Recognition and Image Analysis
- Vol. 36 (2) , 272-284
ShopSafe: Artificial Intelligence-Powered Theft Prevention Using Multiscale Analysis
Shoplifting has become a growing problem for businesses, large and small, whether they are standalone stores or major retail stores. It is usually practiced by people hiding items in the clothes or bags they carry away without paying at the premises. However, no automated approach to this problem is available yet. Therefore, we proposed an artificial intelligence monitoring system that autonomously recognizes shoplifting incidents in retail environments by identifying object-stealing actions in video sequences to address this issue. When such events are detected, the system generates a notification. Our approach proposes a multiscale parallel feature aggregation network to extract salient features in different video sequences. Then, a long short-term memory model was combined to categorize activities and detect shoplifting events. This work makes a significant contribution by optimizing the recurrent learning classifier parameters in relation to sequence length and batch size. We also made available a novel shoplifting dataset, created in a lab-style environment and containing synthesized video clips of normal and shoplifting activities. The proposed system obtained 91.48% accuracy, surpassing existing methods and highlighting its potential as a robust and effective solution for shoplifting detection.
- Open Access
- 21 May 2026
- Published by Oriental Scientific Publishing Company in Biomedical and Pharmacology Journal
- Vol. 01_Special (19) , 1159
Hybrid Approach for Classification and Segmentation of Brain Tumour
Brain tumours pose a significant health challenge, necessitating precise classification and segmentation for effective diagnosis and treatment planning. Manual MRI interpretation is time-consuming and prone to subjectivity, highlighting the need for automated, reliable solutions. This paper presents a hybrid model that combines EfficientNetB7 and UNet to provide tumour classification and segmentation inside a single framework, thereby improving diagnostic accuracy and processing efficiency. EfficientNetB7 extracts high-level features from 600*600*3 MRI scans. This facilitates the precise classification into glioma, meningioma, pituitary, and no tumour. The UNet decoder utilises these shared features to generate pixel-wise segmentation maps of the tumour core, edema, and enhancing tumour regions, with the latter being clinically relevant primarily for gliomas. When tested on the BraTS 2020 dataset, the model achieved a classification accuracy of 98.5%, a Dice coefficient of 94.5%, and an Intersection over Union (IoU) of 91.3%, outperforming standalone EfficientNet, UNet, and contemporary hybrid architectures. With a 42 ms inference time per image, the model enables real-time clinical applications. By combining segmentation and classification into a single framework, the EfficientNetB7-UNet model offers a reliable, effective, and practically useful method for automated brain tumour diagnosis.
- 21 May 2026
- Published by Oriental Scientific Publishing Company in Biomedical and Pharmacology Journal
- Vol. 01_Special (19) , 1159
Open AccessHybrid Approach for Classification and Segmentation of Brain Tumour
Brain tumours pose a significant health challenge, necessitating precise classification and segmentation for effective diagnosis and treatment planning. Manual MRI interpretation is time-consuming and prone to subjectivity, highlighting the need for automated, reliable solutions. This paper presents a hybrid model that combines EfficientNetB7 and UNet to provide tumour classification and segmentation inside a single framework, thereby improving diagnostic accuracy and processing efficiency. EfficientNetB7 extracts high-level features from 600*600*3 MRI scans. This facilitates the precise classification into glioma, meningioma, pituitary, and no tumour. The UNet decoder utilises these shared features to generate pixel-wise segmentation maps of the tumour core, edema, and enhancing tumour regions, with the latter being clinically relevant primarily for gliomas. When tested on the BraTS 2020 dataset, the model achieved a classification accuracy of 98.5%, a Dice coefficient of 94.5%, and an Intersection over Union (IoU) of 91.3%, outperforming standalone EfficientNet, UNet, and contemporary hybrid architectures. With a 42 ms inference time per image, the model enables real-time clinical applications. By combining segmentation and classification into a single framework, the EfficientNetB7-UNet model offers a reliable, effective, and practically useful method for automated brain tumour diagnosis.
- Open Access
- 20 March 2026
- Published by Oriental Scientific Publishing Company in Biomedical and Pharmacology Journal
- Vol. 1 (19) , 246-263
Early-Stage Diabetes Prediction Using a Stacked Ensemble Model Enhanced with SHAP Explainability
Diabetes is one of the most prevalent diseases of our time, and, untreated, it can lead to other health issues. The objective of this research paper is to develop an explainable stacked ensemble model for the early diagnosis of diabetes. The Early-Stage Diabetes Risk Prediction dataset was preprocessed using mean imputation, SMOTE-based class balancing, and mean normalization. A stratified train–test split was applied, and a stacked ensemble model was developed, utilising SHAP and LIME to ensure explainable and interpretable predictions. The proposed model achieved higher performance regarding the Early Stage Diabetes Risk Prediction dataset than did typical models, including Naive Bayes (NB), k-Nearest Neighbour (KNN), Support Vector Machine (SVM), and Decision Tree (DT), with an accuracy of 98.4%. The innovative application of ensemble learning enhances the model's reliability and effectiveness, offering valuable insights for identifying potential diabetic patients. The high accuracy underscores the model's potential as a valuable tool for early detection and treatment, ultimately improving patient outcomes in diabetes management. A critical aspect of our methodology is the integration of SHAP (SHapley Additive exPlanations) and Local Interpretable Model-Agnostic Explanations (LIME), which enhances explainability by revealing the factors driving the model's predictions and highlighting feature importance.
- 20 March 2026
- Published by Oriental Scientific Publishing Company in Biomedical and Pharmacology Journal
- Vol. 1 (19) , 246-263
Open AccessEarly-Stage Diabetes Prediction Using a Stacked Ensemble Model Enhanced with SHAP Explainability
Diabetes is one of the most prevalent diseases of our time, and, untreated, it can lead to other health issues. The objective of this research paper is to develop an explainable stacked ensemble model for the early diagnosis of diabetes. The Early-Stage Diabetes Risk Prediction dataset was preprocessed using mean imputation, SMOTE-based class balancing, and mean normalization. A stratified train–test split was applied, and a stacked ensemble model was developed, utilising SHAP and LIME to ensure explainable and interpretable predictions. The proposed model achieved higher performance regarding the Early Stage Diabetes Risk Prediction dataset than did typical models, including Naive Bayes (NB), k-Nearest Neighbour (KNN), Support Vector Machine (SVM), and Decision Tree (DT), with an accuracy of 98.4%. The innovative application of ensemble learning enhances the model's reliability and effectiveness, offering valuable insights for identifying potential diabetic patients. The high accuracy underscores the model's potential as a valuable tool for early detection and treatment, ultimately improving patient outcomes in diabetes management. A critical aspect of our methodology is the integration of SHAP (SHapley Additive exPlanations) and Local Interpretable Model-Agnostic Explanations (LIME), which enhances explainability by revealing the factors driving the model's predictions and highlighting feature importance.
- Open Access
- 16 March 2026
- Published by Sakarya University Journal of Computer and Information Sciences in Sakarya University Journal of Computer and Information Sciences
- Vol. 9 (1) , 190-204
ViT-HAR: Vision Transformer-Based Human Activity Recognition in Cluttered Environments
In the real world, Human Activity Recognition (HAR) remains challenging due to issues such as occlusion, dynamic backgrounds, and visual noise. Traditional models, such as CNNs, RNNs, and ST-GCNs, have constraints, including a small receptive field and the use of local features, which also reduce generalisation. We present ViT-HAR, a Vision Transformer framework that learns global spatio-temporal interactions and proposes two new modules, namely Contextual Patch Reweighting (CPR) and Attention-Guided Occlusion Masking (AGOM) to solve this issue. These parts enable selective attention of motion-relevant and nonoccluded areas, which increases robustness and interpretability in cluttered scenarios. In contrast to the previous versions of Vision Transformer architecture (ex, TimeSformer and ViViT), which use fixed attention by default, ViT-HAR is based on adaptive attention, redistributing contextual patches and masking unseen areas with dynamically varying weights in attempts to retain semantically salient information. The combined pipeline utilises dynamic frame sampling, contextual reweighting, and occlusion-based masking, resulting in an optimal trade-off between spatial and temporal coherence. NTU RGB+D and Kinetics-700 and UCF101 evaluation results indicate a maximum 6.5% greater Top-1 and show better F1-scores than 3D-CNN and RNN hybrids. Visualization of attention maps attests to the fact that ViT-HAR pays attention to meaningful motion signals, which is why the algorithm proves useful in healthcare monitoring, smart surveillance, and AR/VR. Lightweight and multimodal extensions are investigated in future work.
- 16 March 2026
- Published by Sakarya University Journal of Computer and Information Sciences in Sakarya University Journal of Computer and Information Sciences
- Vol. 9 (1) , 190-204
Open AccessViT-HAR: Vision Transformer-Based Human Activity Recognition in Cluttered Environments
In the real world, Human Activity Recognition (HAR) remains challenging due to issues such as occlusion, dynamic backgrounds, and visual noise. Traditional models, such as CNNs, RNNs, and ST-GCNs, have constraints, including a small receptive field and the use of local features, which also reduce generalisation. We present ViT-HAR, a Vision Transformer framework that learns global spatio-temporal interactions and proposes two new modules, namely Contextual Patch Reweighting (CPR) and Attention-Guided Occlusion Masking (AGOM) to solve this issue. These parts enable selective attention of motion-relevant and nonoccluded areas, which increases robustness and interpretability in cluttered scenarios. In contrast to the previous versions of Vision Transformer architecture (ex, TimeSformer and ViViT), which use fixed attention by default, ViT-HAR is based on adaptive attention, redistributing contextual patches and masking unseen areas with dynamically varying weights in attempts to retain semantically salient information. The combined pipeline utilises dynamic frame sampling, contextual reweighting, and occlusion-based masking, resulting in an optimal trade-off between spatial and temporal coherence. NTU RGB+D and Kinetics-700 and UCF101 evaluation results indicate a maximum 6.5% greater Top-1 and show better F1-scores than 3D-CNN and RNN hybrids. Visualization of attention maps attests to the fact that ViT-HAR pays attention to meaningful motion signals, which is why the algorithm proves useful in healthcare monitoring, smart surveillance, and AR/VR. Lightweight and multimodal extensions are investigated in future work.
- 29 January 2026
- Published by IGI Global Scientific Publishing
- p. 401-434
Nutrition Meets AI
This chapter examines the personalization of nutrition using AI-powered recommender systems, focusing on how they can improve health goals and encourage sustainable eating habits. It incorporates the latest developments in machine learning, including matrix factorization, convolutional neural networks, reinforcement learning, and graph neural networks in devising automated food identification, meal planning, and interactive dietary recommendations. The chapter describes how these systems, using biometric data like glucose levels and gut microbiome profiles, behavior analytics such as eating and preference patterns, and contextual information like time, location, and cultural understanding, provide personalized recommendations tailored to individuals' unique health profiles. The practical applications range from dealing with chronic illnesses and obesity to sensitive dieting, demonstrating their practicality. Actionable recommendations are derived from the comparative analysis of benchmark datasets Nutrition5k, MyFoodRepo, and FoodKG, as well as the models' performance evaluations.
- 29 January 2026
- Published by IGI Global Scientific Publishing
- p. 401-434
Nutrition Meets AI
This chapter examines the personalization of nutrition using AI-powered recommender systems, focusing on how they can improve health goals and encourage sustainable eating habits. It incorporates the latest developments in machine learning, including matrix factorization, convolutional neural networks, reinforcement learning, and graph neural networks in devising automated food identification, meal planning, and interactive dietary recommendations. The chapter describes how these systems, using biometric data like glucose levels and gut microbiome profiles, behavior analytics such as eating and preference patterns, and contextual information like time, location, and cultural understanding, provide personalized recommendations tailored to individuals' unique health profiles. The practical applications range from dealing with chronic illnesses and obesity to sensitive dieting, demonstrating their practicality. Actionable recommendations are derived from the comparative analysis of benchmark datasets Nutrition5k, MyFoodRepo, and FoodKG, as well as the models' performance evaluations.