An Efficient Early-Onset Plant Disease Prediction using Improved Heuristic-aided Adaptive Ensemble Network with Leaf Image-based Phenotype Data

Plant-related diseases pose a pressing threat to the agricultural industry, which is already strained to meet growing food demands. Farmers, whose primary income relies on agricultural production, often confront significant challenges as these diseases can severely disrupt crop growth and quality. Without early prediction, such diseases can greatly reduce crop productivity. Hence, to overcome this threat at an early stage, this research concentrates on developing an effective Adaptive Ensemble Network with a novel loss and activation function model for early-onset plant disease prediction using plant leaf imagery. In this method, the vital features are extracted from the Residual Network (ResNet152), Visual Geometry Group (VGG19), and DenseNet161. After attaining the features, the final feature set is obtained by the averaging-based computation. Then, the resultant features are given to the Deep Temporal Convolution Network (DTCN) for prediction of early-onset plant disease, in which the loss and activation functions are newly derived. Furthermore, parameter tuning uses the Modified Update in the Coati Optimization Algorithm (MUCOA). Finally, the validation outcomes of the designed approach are validated against conventional frameworks. The suggested framework performs robustly, achieving maximum accuracy, sensitivity, and specificity values exceeding 93% across both datasets. As a result, the proposed AEN model can be an insightful, farmer-friendly aid in identifying and predicting the early beginnings of plant leaf diseases.