Plant Disease Recognition from the Perspective of Bangladesh: A Comparative Study of Deep Learning Models and Ensemble Techniques

Plant disease detection is one of the most crucial tasks in modern agriculture, and early detection is essential to minimize crop loss. In this work, we evaluate several deep learning models, including Xception, VGG-16, VGG-19, ResNet-101V2, ResNet-152V2, and ConvNeXtBase, for plant disease detection. We also propose an ensemble learning approach by combining the outputs of Xception, VGG-19, and ResNet-152V2 to improve prediction results. For each model, performance is evaluated using metrics such as accuracy, F1 score, precision, recall, and ROC-AUC. Among the individual models, the ensemble model achieved a test accuracy of 95.45%, a validation accuracy of 94.51%, an F1 score of 95.27%, and an ROC-AUC of 99.93%. These results demonstrate that ensemble methods are highly effective in leveraging the strengths of diverse architectures, improving both classification accuracy and robustness. This approach proves to be particularly effective for real-world applications in plant disease management.