A novel approach to detect tomato leaf disease using vision transformer

Abstract

Tomatoes are one of the most widely consumed vegetables across the world. However, tomatoes are prone to diseases. Recognizing and classifying tomato leaf diseases is crucial task. Various deep learning (DL) methods have been developed by several researchers, but they have some complex issues like noise in images, high computational complexity, poor accuracy, and limited feature selection. The main goal of this research is to present novel DL based tomato leaf disease classification framework with neural network based gated vision transformer (G-ViT) model assisted attention mechanism. The proposed framework uses dilated convolution with bidirectional long short-term memory (Bi-DLSTM) used for efficient feature extraction to enhance the classification. An effective chaotic spider wasp optimization (CSWO) is used for feature selection. Further, novel attention based gated vision transformer (A-GVT) is used to classify tomato leaf diseases which integrates strengths of attention mechanism and G-ViT models. Further, to improve the generalizability of classification model, its parameters are tuned with black widow optimization (BWO) algorithm. The experimental findings shows that proposed framework outperformed previous studies on tomato leaf disease identification and classification models in terms of accuracy, precision, recall, F1-score, specificity, mean absolute error (MAE), and root mean square error (RMSE) with 99.7%, 98.29%, 98.22%, 98.25%, 99.19%, 0.03, and 0.25 respectively. The proposed study can pave a way for new agricultural revolution.