Optimizing CNN Performance for Tomato Disease Classification with Advanced Data Augmentation Techniques

Abstract

Tomato (Solanum lycopersicum), a vital horticultural crop, faces increasing challenges from diseases such as bacterial spot,
tomato mosaic, and yellow leaf curl, causing substantial global crop losses. Timely and accurate disease detection is crucial
for effective management strategies. This research introduces a sophisticated method for detecting tomato leaf diseases by
enhancing a model with diverse data augmentation techniques. Evaluation metrics including precision, recall, and F1-score
consistently demonstrate high performance, ranging from 1.00 to 0.99 to 1.00, respectively. By incorporating Mixup,
CutMix, Adversarial examples, Style Transfer, and Image Blending during training, the model achieves remarkable
validation accuracies: 99.47%, 99.26%, 99.42%, 99.68%, and 99.63%, respectively. Notably, the highest accuracy of
99.68% is achieved using Style Transfer augmentation. In contrast, a Convolutional Neural Network (CNN) employing
conventional augmentation techniques achieves a prediction accuracy of 98.99% for the same tomato diseases. These results
underscore the significant improvement in disease prediction accuracy through the integration of advanced augmentation
techniques with CNNs. The study highlights CNNs with advanced augmentation as the optimal choice for accurately
predicting tomato diseases.