My Research, Patents & CertificationsMy Research, Patents & Certifications

This research focuses on developing attention-enhanced deep learning models, including Conv-LSTM and Bi-LSTM, to assess and forecast groundwater quality using physicochemical Q-values. Groundwater samples collected from the Russell River, Australia, between December 2016 and April 2020 (approximately 1,300 datapoints) were analyzed to parameterize Q-values, a standardized measure for comprehensive water quality assessment. The proposed models provide a distinctive framework for forecasting Q-values and supporting timely decision-making in water resource management. Among the models, the Bidirectional LSTM with Attention achieved the highest predictive performance, with a root mean square error of 0.0057, a mean absolute error of 0.0022, a symmetric mean absolute percentage error of 3.8875%, and a coefficient of determination of 0.9910. This work demonstrates a reliable, scalable, and effective decision-support framework for sustainable water resource management and policy-making.

This research focuses on developing a hybrid model using Graph Attention Network (GAT) and Graph Convolution Network (GCN) to improve leaf disease classification in plants. Superpixel segmentation was employed for efficient feature extraction, partitioning images into homogeneous regions to better capture localized features, while edge augmentation was used to enhance model generalization. The model was evaluated on apple, potato, and sugarcane leaves, and achieved precision, recall, and F1-scores of 0.9822, 0.9818, 0.9818 for apple; 0.9746, 0.9744, 0.9743 for potato; and 0.8801, 0.8801, 0.8799 for sugarcane leaf disease classification, respectively. The results demonstrate that the GAT-GCN hybrid model effectively and consistently identifies leaf diseases, providing a robust framework for timely agricultural disease monitoring and management.

This research introduces LeafNet, a deep learning architecture designed for detecting six major groundnut leaf classes, including Early Rust, Early Leaf Spot, Nutrition Deficiency, Rust, Late Leaf Spot, and Healthy Leaf, using a dataset of 10,361 images. The model incorporates residual networks and weight initialization techniques to improve robustness and training efficiency. LeafNet was evaluated against state-of-the-art neural network architectures and achieved a test accuracy of 97.225%, precision of 97.365%, recall of 97.225%, F1-score of 97.225%, and MCC of 96.700%. Additional testing on diverse leaf disease datasets confirmed its adaptability and generalizability. This work contributes to agricultural science by providing a reliable and accurate deep learning solution for timely disease detection and sustainable crop management.