TomSevNet: a hybrid CNN model for accurate tomato disease identification with severity level assessment

Tomato diseases are a major challenge for tomato growers, leading to significant yield losses and reduced quality of produce. Manual diagnosis of tomato diseases can be time-consuming and error-prone, making automated disease diagnosis an attractive solution. In this work, we develop a hybrid convolutional neural network (CNN) model using self-regulated layers and inception layer named as TomSevNet (Tom-Tomato disease Sev-Severity Net-Network) for accurate and efficient diagnosis of tomato diseases with severity levels. Our approach involves training a TomSevNet model on Plant Village dataset of tomato diseases segregated into different categories with their severity levels. The TomSevNet model is trained on a dataset containing 30 different classes belonging to nine tomato diseases of three severity levels, a healthy class, and two negative classes. Negative classes are included in the dataset to avoid misclassification problem. The TomSevNet classifier with Adadelta optimizer has performed extremely well and has attained the highest testing accuracy of 96.91% and the F1-score is 0.97. We also performed a comparison with other bench marked reference models, and the TomSevNet model outperformed them in terms of accuracy as well as F1-score.