S. Priya

In the agricultural sector, pistachio classification is crucial for inventory management, quality assurance, and sorting, as it is necessary to differentiate between several pistachio varieties. Deep learning techniques, in particular convolutional neural networks (CNNs), have demonstrated promising results in a variety of areas in image classification tasks in recent years. This work proposes Kernel Isolated Pyramid Pooled VGG16 (KIP-VGG16) to address the multi scale feature pooling of pistachio image for its type classification. This research utilizes the images from Pistachio Image KAGGLE public repository dataset. The initial contribution of this work was the formation of Kernel isolated pistachio image that reduces the interference from background noise and irrelevant features. The second contribution deals with refining the existing VGG16 by replacing the max pooling in the convolution block to pyramid pooling to propose KIP-VGG16 model. The kernel-isolation filtering technique was used that concentrated on extracting the multi-scale features, enabling for more robust feature extraction from images of pistachios under varying conditions. Pyramid pooling was adapted inside the convolution that captures both fine-grained details and global context, ensuring effective representation across varying image sizes and orientations. The KIP-VGG16 model can be applied for tasks where multiple pistachio types need to be identified from complex backgrounds and varying angles. The filtered pistachio images are divided in the ratio of 80: 10:10 and are executed with the CNN and KIP-VGG16 model to assess the performance. Implementation shows that KIP-VGG16 found to exhibit high accuracy of 99.47% towards the classification of Kirmizi and Siirt pistachio classes. This novel KIP-VGG16 method creates new opportunities for high-performance image classification tasks in agriculture, where accurate pistachio type sorting and identification are essential for meeting consumer requests and ensuring product quality.

Deep learning-based nut classification has emerged as a viable way to automate the detection and categorization of different nut varieties in the food processing and agriculture sectors. Conventional techniques for classifying nuts mostly rely on manually created characteristics like texture, color, shape, or edges. These characteristics frequently fall short of capturing the image’s complete complexity, particularly when nuts show tiny visual variances. This research proposes Deep Atrous Context Convolution Generative Adversarial Network (DAC-GAN) model that categorize the 8 classes of nuts like brazil nuts, cashew, peanut, pecan nut, pistachio, chest nut, macadamia and Walnut. This research uses Common Nut KAGGLE dataset with 4,000 nuts images of 8 nuts classes. The DAC-GAN approach overcomes the difficulties of having limited labelled data for nut classification tasks by employing DCGANs’ ability to produce high-quality, synthetic nut images to supplement the dataset. The DCGAN comprises of a discriminator and a generator block. The discriminator block develops the ability to differentiate between synthetic and real images, while the generator block generates realistic nut images from random noise. The real images along with the DCGAN generated images are processed with feature filtering methods to extract the Corner Key Points Featured (CKPF) nuts images. To further enhance the feature selection, the CKPF edges are extracted from the image that provides unique, geometrically distinctive critical corners to further process for representative learning. To proceed with the effective feature extraction and model learning, the CKPF nuts images are processed with atrous convolution that capture the intricate details by expanding the receptive field without losing resolution. The novelty of this work exists by appending the filtration and atrous convolution that acquire the spatial data features from the nut’s images at various resolutions. Atrous convolution was refined by appending the pre-context and post-context block that add the image level information to the features. The effectiveness of the DAC-GAN model was validated with the traditional augmented dataset with all existing filtering images and CNN models. Implementation outcome shows that DAC-GAN found to exhibit high accuracy of 99.83% towards the nuts type classification. The superiority of the DAC-GAN method over traditional approaches is demonstrated by extensive experiments on augmented and DCGAN generated datasets, which achieve higher classification accuracy and generalization across a variety of nut type categorization. The outcome demonstrates that the DCGAN together with atrous convolution have the potential to be an effective tool for automating nut sorting in food industry.

Malware detection remains challenging because of privacy constraints, heterogeneous data distributions, and vulnerability to adversarial attacks in distributed environments. We proposed a federated adversarial learning framework for robust, privacy‐preserving malware classification. The framework introduces a two‐way federated strategy, in which two client groups train vision transformers using different patch sizes to capture fine‐ and coarse‐grained malware features. A hierarchical selective aggregation mechanism enhances the global model by weighting trustworthy client contributions and improving resilience to stragglers and heterogeneous data, while integrated adversarial training mitigates model poisoning and data leakage. Experiments conducted on the Malevis, BIG2015, and Malimg benchmarks demonstrated strong scalability, achieving high accuracy with four clients, and minimal performance degradation at eight and 12 clients. The proposed approach further outperformed standard and adversarial attention‐based vision transformers as well as state‐of‐the‐art federated approaches, delivering a scalable, secure, and interpretable solution for modern malware detection.

Mouth cancer represents a significant global public health challenge due to its high incidence rate and potentially fatal outcomes if not diagnosed early. Among the various types, oral cancer is notably prevalent and poses considerable diagnostic complexity due to its intricate histopathological architecture. According to the World Health Organization and recent epidemiological studies, oral cancer accounts for approximately 377,000 new cases and 177,000 deaths annually worldwide. Traditional diagnostic methodologies such as manual clinical inspection and biopsy analysis are often time-intensive, inherently subjective, and susceptible to inter-observer variability among pathologists. To address these limitations, this study proposes novel framework, termed Depthwise Separable Convolution U-Net (DWSU-Net), aimed at enhancing the accuracy and efficiency of mouth cancer detection. Unlike traditional U-Net, which employs computationally expensive standard convolutional layers, DWSU-Net integrates depthwise separable convolutional blocks into both encoder and decoder stages, reducing parameter count and training complexity while preserving representational power. This makes the model more lightweight, scalable, and suitable for real-time or resource-constrained clinical environments. The research utilizes histopathologic images of oral tissues obtained from the publicly available oral cancer detection dataset on Kaggle, which were captured using a Leica ICC50 HD microscope. The proposed approach initiates with a comprehensive data preprocessing pipeline involving multiple filtering techniques. Raw histopathological images are transformed using Sobel filtering, Otsu thresholding, Canny edge detection, and morphological reconstruction via erosion, thereby improving feature saliency and contrast in malignant regions. The preprocessed dataset is partitioned into training, validation, and testing subsets in an 80:10:10 ratio and evaluated using fivefold cross-validation to ensure the robustness and generalizability of the model. A comparative analysis was conducted using conventional CNN architectures to identify the most effective combination of model and filtering technique. Empirical results indicated that MobileNet and U-Net, when applied to images filtered through morphological reconstruction by erosion, yielded superior classification performance. Motivated by these findings, the proposed DWSU-Net architecture was developed by integrating the strengths of MobileNet and U-Net. The novelty of the DWSU-Net model lies in morphological reconstruction-based preprocessing, which enhances interpretability by making malignant regions more distinct for both the algorithm and pathologists, thereby bridging the gap between AI predictions and clinical understanding. By combining efficiency, robustness, and interpretability, DWSU-Net goes beyond accuracy gains, offering a clinically relevant decision-support framework. Experimental evaluations demonstrate that the proposed model achieves an outstanding classification accuracy of 99.74% in distinguishing between healthy and malignant oral tissue images, underscoring its potential for clinical deployment in automated cancer diagnostics.

Early identification of root diseases is vital for safeguarding crop health and maximizing agricultural yield. Traditional diagnostic approaches are often constrained by their manual nature, slow response time, limited scalability, and inconsistent accuracy. To overcome these limitations, this study proposes a novel hybrid RIFATA Attention-PSPNet (RA-PSPNet) framework designed for automated and precise root disease identification. The RA-PSPNet replaces the CNN to Convolutional block attention module (CBAM) in the Pyramid Scene Parsing Network (PSPNet). Further, RA-PSPNet is optimized with a metaheuristic optimization technique called the Remora Improved Feedback Artificial Tree Algorithm (RIFATA), which enhances model performance through dynamic hyperparameter tuning. The proposed system includes a multi-stage pipeline: root images are first pre-processed using Rot Sensitive Gaussian (RSG) filtering, followed by processing with CBAM to provide multiscale attention feature map and finally segmented PSPNet. Both the segmentation and classification components of proposed RA-PSPNet are optimized using RIFATA. RIFATA combines the strengths of the Remora Optimization Algorithm (ROA) and Improved Feedback Artificial Tree Algorithm (IFATA), which incorporates elements from Improved Invasive Weed Optimization (IIWO) and Feedback Artificial Tree (FAT) models for efficient hyperparameter tuning. Experimental evaluations conducted on multiple datasets from Maize root, Root cowpea, Wheat root, Rice root gellan, Alfalfa root, Chrono root, Root crown images of soybean and wheat datasets in the ratio of 80:10:10. Implementation reveals that proposed RA-PSPNet outperforms with the accuracy of 99.63%, high sensitivity (99.13%), specificity (99.23%), precision (99.03%), recall (99.13%), and F1-score (99.08%) towards healthy and rot root classification. These results demonstrate the effectiveness and scalability of RA-PSPNet for sustainable precision agriculture and root disease diagnosis. The implementation are documented in our publicly accessible repository on Zenodo to enhance the transparency and reproducibility. (https://doi.org/10.5281/zenodo.16825955).

Partial shading poses a significant challenge in photovoltaic systems by creating multiple peaks in the power-voltage curve, complicating the task of accurately tracking the Maximum Power Point. Traditional maximum power point tracking methods often struggle to identify the true Global Maximum Power Point, leading to suboptimal energy harvesting. This paper proposes a novel hybrid tracking framework that integrates fuzzy logic, synthetic data generation using Generative Adversarial Networks (GANs), and time-series modeling with Transformer architectures. Fuzzy logic improves resilience to input uncertainties by translating raw data into interpretable fuzzy values. GANs augment the dataset by generating realistic synthetic samples, thereby improving generalization. The Transformer model leverages self-attention mechanisms to capture long-term temporal patterns in solar irradiance and power profiles. By combining these strengths, the proposed method delivers a robust and accurate global maximum power point tracking solution, particularly under dynamic and partially shaded environments. Experimental results demonstrate its superior performance and scalability compared to conventional maximum power point tracking approaches.

Kidney stone classification is a critical yet complex task in medical imaging, traditionally performed using computed tomography (CT) and ultrasound scans. Manual interpretation of these images is time‐consuming and prone to variability, highlighting the need for automated diagnostic solutions. This study proposes Contour‐Detected Normalized Residual VGG19 (CDR‐VGG19), a deep learning model inspired by VGG19 and enhanced with residual connections to improve classification accuracy. The model leverages contour detection for unsupervised feature extraction, followed by supervised learning using a hybrid of VGG19 and ResNet architectures. Using the Kidney Stone KAGGLE dataset of 2602 images, the model applies data augmentation, preprocessing, and feature filtering. Images are split into training, validation, and testing sets (80:10:10), and multiple CNNs are evaluated. Results show that the proposed CDR‐VGG19 achieves a high classification accuracy of 99.61%, demonstrating its effectiveness in detecting kidney stones from contour‐enhanced images.