MSNet-LNet Architecture With Improved Deep Joint Segmentation for Tomato Plant Disease Classification With Set of Multi-Texton and Statistical Features

Abstract

Tomato is a broadly consumed crop in every continent, though the number of tomatoes varies based on the method of fertilisation. The main factor that affects the crop yield quantity and quality is leaf disease. Therefore, it is essential to correctly diagnose and categorise these illnesses. Many creative approaches have been proposed to diagnose and categorise diseases using images. The major objective is to present a five-phase tomato plant disease classification (TPDC) model. Initially, Gaussian filtering is deployed as a pre-processing technique on the input image. To perform the segmentation procedure, an improved deep joint (IDJ) framework is suggested. Next, features are retrieved, such as statistical features, colour features and Improved MT (IMT) features. Data augmentation is then carried out. Based on the training feature information, a hybrid model integrating the modified Squeeze Net (MSNet) and LinkNet is presented to perform disease classification on tomato plants. Moreover, the proposed method is validated using two benchmark datasets, namely the PlantVillage dataset and the Taiwan tomato leaf images. The performance of the proposed work is evaluated over the conventional models in terms of different performance measures. As a result, the proposed IHM (MSNet+Linknet) model surpassed traditional methods, achieving an accuracy of 0.957, specificity of 0.975 and NPV of 0.973, consistently showcasing strong performance across both datasets. Therefore, the proposed approach shows exceptional performance in classifying tomato leaf diseases over the traditional methods.