Effective Tomato Leaf Disease Identification Model using MobileNetV3Small

Abstract

Agriculture is widely affected by various diseases. The traditional methods of detecting diseases are time-consuming and labor-intensive. Deep learning is one of the most emerging fields. It has largely evolved in recent years because of its automatic feature extraction mechanism. Deep learning in agriculture has shown promising results for analyzing various agricultural aspects. Because of the automated methods deep learning has shown very promising results in finding leaf diseases automatically without requiring human intervention. There are many crops which play a big role in the economy of any county. One of the vegetables that is most frequently used in cooking is the tomato. The tomato crop is more vulnerable to diseases. The diseases in the tomato crop cause significant damage to the crop. It is difficult to identify the disease early because of many constraints. Agronomists and farmers can more efficiently and promptly manage the crop with the use of a deep learning-based method for detecting tomato leaf diseases. In this research, we have trained various deep learning models on the publically available dataset of tomatoes which contained 9 leaf diseases and 1 healthy. The models used are VGG16, MobileNetV2, SqueezNet and MobileNetV3Small. Different models provided a different accuracy on the public dataset. MobileNetV3Small model showed a very promising accuracy of 99.43% on 22930 leaf images of tomatoes. The accuracy of the model achieved on 25 epochs.