Development of a Tomato Leaf Disease Detection System Using Convolutional Neural Networks

Abstract

The study and early detection of plant diseases are important because they harm plants and essential products like food, clothing, furniture, and housing. Plant diseases such as late blight, early blight, and septoria leaf spot significantly reduce tomato crop yields, posing major challenges to sustainable agriculture and food security. This paper aims to develop a cloud-based software system to accurately diagnose and recommend diseases in tomato leaves from images, providing recommended treatments for the detected diseases. The system integrates three neural networks: two Convolutional Neural Networks for object detection and feature extraction, and a third neural network for classification. These networks were trained on images from the Plant Village dataset, with preprocessing to enhance image quality and annotation accuracy. The system was tested and verified, achieving a training accuracy of 98% and a loss of less than 7.6% on the test set. However, real-world testing indicated an accuracy of around 80% for non-test-set images. The developed system has the potential to significantly aid in the early detection and management of tomato plant diseases, improving crop yields and agricultural sustainability. The software is accessible via an Application Programming Interface developed using TypeScript, and a frontend web application serves as a human-machine interface.