Transfer Learning VGG16 Model for Classification of Tomato Plant Leaf Diseases: A Novel Approach for Multi-Level Dimensional Reduction

Abstract

Tomato is the most popular and cultivated crop in the world. Nevertheless, the quality and quantity of tomato crops have been declining due to various diseases that afflict tomato crops. Hence, it becomes necessary to detect the disease early to prevent crop damage and increase the yield. The proposed model in this article predicts the infected tomato leaf images (9 classified diseases and also healthy class) obtained from the Plant Village dataset. In this model, Transfer learning was used to extract features from images by VGG16, yielding a high dimension of 25088 features. Overfitting is a commonly anticipated problem because of the higher dimensionality of data. To mitigate this problem, the authors have adopted a novel dimensional reduction-based technique: filter methods, feature extraction techniques like Principal Components Analysis (PCA), and the Boruta feature selection technique of wrapper methods. This adoption enables the proposed model to attain a significantly improved high accuracy of 95.68% and 95.79% in MLP and VGG16, respectively, by reducing its initial dimension on the tomato dataset containing 18160 images across 10 classes.