HybOptic-CNN: A hybrid WOA-GWO-optimized convolutional neural network model for enhanced plant disease detection in the Nigerian environment

Abstract

Plant diseases threaten agricultural productivity, and automated image analysis can support early identification of visible disease symptoms. This study introduces HybOptic-CNN, a convolutional neural network (CNN) whose learning rate and batch size are selected using a hybrid Whale Optimization Algorithm--Grey Wolf Optimizer (WOA-GWO). Nine disease classes were selected from the 22-class CCMT field-image dataset, and 152 local farm leaf images were collected in Enugu State, Nigeria. Of the local images, 122 (80.3%) were added to the model-development data for training and validation, whereas 30 (19.7%) formed an independent Nigerian hold-out set excluded from augmentation, class balancing, early stopping, validation, and hyperparameter selection. Across 10 model-development runs, the optimized model achieved 96.8 ± 0.4% mean validation accuracy, 95.2 ± 0.5% macro-precision, 94.9 ± 0.6% macro-recall, and 95.0 ± 0.5% macro-F1, compared with 90.3 ± 0.9% validation accuracy and 86.3 ± 1.2% macro-F1 for the baseline. The optimized model improved mean validation accuracy by 6.5 percentage points and converged 14.6 epochs earlier. On the independent 30-image Nigerian hold-out, HybOptic-CNN achieved 93.3% accuracy and 93.1% macro-F1 across four represented disease classes. A web application integrating the trained classifier was also demonstrated. These results support improved model-development performance through hybrid hyperparameter selection and motivate broader multi-location field evaluation.