Tomato leaf diseases recognition based on deep convolutional neural networks

Abstract

Tomato disease control remains a major challenge in the agriculture sector. Early-stage recognition of these diseases is critical to reduce pesticide usage and mitigate economic losses. While many research works have been inspired by the success of deep learning in computer vision to improve the performance of recognition systems for crop diseases, few of these studies optimised the deep learning models to generalise their findings to practical use in the field. In this work, we proposed a model for identifying tomato leaf diseases based on in-house data and public tomato leaf image databases. Three deep-learning network architectures (VGG16, Inception_v3, and Resnet50) were trained and tested. We packaged the trained model into an Android application named TomatoGuard to identify nine kinds of tomato leaf diseases and healthy tomato leaves. The results showed that TomatoGuard could be adopted as a model for identifying tomato diseases with a 99% test accuracy, showing significantly better performance than APP Plantix, a widely used APP for general-purpose plant disease detection.