Comparative Performance of Deep Learning Algorithms in Tomato Disease Detection

Agriculture is crucial for the economies of emerging nations, providing subsistence, employment, and livelihoods for rural communities. However, crop diseases pose a risk to food security, income, and human health necessitating timely identification and appropriate treatments. A disease can slow plant growth, reduce productivity, and even result in crop failures. Therefore, it is crucial to develop sustainable agricultural methods and technological advances in crop production in order to achieve long-term economic and ecological benefits. Tomato is a vital crop in India, affected by diseases such as early blight, late blight, septoria leaf spot, bacterial spot, and yellow leaf curl virus, the symptoms of which are visible on the leaves as different patterns, spots, and deformities. A deep learning model extracts intricate patterns from leaf images to analyze them effectively by identifying symptoms of disease, such as color changes, spots, and variations in texture, with high accuracy. In this paper, a comprehensive comparative analysis of five deep learning models, including a custom Convolutional Neural Network (CNN), transfer learning with VGG16 and Xception, DenseNet, and Inception V3, is used in the identification of leaf diseases. A dataset of 11,606 labeled images, divided into 6 classes representing both diseased and healthy leaves were used to assess the effectiveness of the algorithms. Each model was trained for 10 epochs using an 80:20 training-validation split and evaluated using performance metrics such as accuracy, precision, recall, F1-score, and confusion matrices. The results showed that the Transfer Learning with VGG16 outperformed the other models achieving a weighted average $\mathrm{F 1}$-score of ${0. 9 6}$. Future work will focus on optimizing these models for real-time deployment on edge devices and expanding the dataset to encompass a wider range of diseases and environmental conditions.