Deep Learning Based Tomato Leaves Disease Detection: A Comprehensive Approach Using Convolutional Neural Networks

Diseases of tomato leaves have a high impact on the production of crop and the yield can be reduced if the diseases are not identified at primary stage. The traditional way of detecting diseases by human experts is slow and laborious, and also less accurate in many cases, particularly for large farms. To address this issue, the paper proposes a deep learning based tomato leaf disease detection system utilizing the VGG19 model. The Plant Village dataset, which has images of healthy and infected tomato leaves, was utilized for both training and testing. Prior to training, images were resized, normalized and augmented with rotation, flipping and scaling to enhance model robustness and performance in various testing conditions. This model is trained on 80-10-10 split for training, validating and testing. To enhance learning speed, the categorical cross-entropy loss function and the Adam optimizer were employed. Results of experiments indicate the proposed VGG19 based model is superior enough, achieving the validation accuracy of 91.56% and the test accuracy of 98.36%. Training and validation plot illustrate the model learns well with little overfitting. The low test loss of 0.07 also suggests a good generalization on unseen data. This work demonstrates that deep learning can be used to generate fast, accurate and reliable solutions for tomato leaf disease detection. The proposed technique can aid the farmers in several ways for example detection of disease, preventing the damage to the crop, thereby enhancing the agricultural production. In addition, the work provides a robust platform for developing real time applications on mobile and IoT platforms.