Tomato Disease Detection to Reduce Food Waste

Abstract

The issue of detecting diseases in tomato leaves plays a major role in agricultural sustainability and food safety. Tomatoes are the second most important fruit crop after potatoes. Currently, almost 90% of edible tomatoes are thrown away, significantly contributing to overall fruit and vegetable waste. This research compares the performance of several Convolutional Neural Network (CNN) with the aim to efficiently detect diseases in tomatoes and reduce food waste. From VGG-16 to EfficientNet-B0, six different architectures have been tested to classify 10 distinct diseases of tomatoes using a subset of the public PlantVillage dataset. The results concluded that the best-performing model is the EfficientNet-B0 architecture, with a 96.04% test accuracy. Future work includes augmenting the dataset with more images using C-GAN and alternative techniques, as well as testing different tricks to improve the current performance. The final aim is to contribute to the design of a successful and robust artificial intelligence tool capable of reducing loss and waste of tomatoes. © 2025 SPIE.