Real-Time Disease Identification of Banana Plants Using Neural Network Models

The consumption of agricultural products is fundamental to human survival. Enhancing agricultural productivity and sustainability necessitates the effective monitoring and nurturing of healthy crops. In this regard, one of the most important study areas is using sophisticated neural network models to identify plant diseases. Four well-known convolutional neural network (CNN) architectures—InceptionV3, ResNet50V2, VGG16, and MobileNet—that use transfer learning for real-time disease identification in banana plants are thoroughly compared in this work. A total of 11 disease variants were classified, and performance metrics were computed for each model. Stratified random sampling was employed to ensure balanced representation of classes during training and validation. Evaluation using Cross-Validation, ROC-AUC, and confusion matrices ensured robust and interpretable classification results. Among the models, ResNet50V2 obtained the maximum accuracy of 99.52% on in-distribution test data, whereas MobileNet reached 90% accuracy on real-time datasets. The results were validated by an agricultural expert, confirming their practical reliability. This study provides useful information about the advantages and disadvantages of each architecture, along with practical suggestions for their implementation in actual agricultural environments to help farmers identify and treat diseases promptly.