Deep Learning for Tomato Leaf Disease Detection for Images Captured in Varying Capturing Conditions

Main Article Content

Haridas D. Gadade, Dr. D. K. Kirange


In deep learning characteristics of data are learned using calculation models consisting of multiple processing layers. It is a form of machine learning technique. I this paper we have employed convolutional neural network models for identification of tomato plant leaf disease using simple leaf images of healthy and diseased plants. Deep learning methodology is used for tomato leaf disease identification for images captured in varying capturing conditions. We have collected the tomato leaf disease dataset containing images with complex background, images containing other leaves and other parts of plants. We have evaluated the performance of deep learning CNN with different machine learning models including color, texture and shape features with SVM, KNN, Naive Bayes, Decision trees and LDA classification. Deep Learning model with CNN gives better accuracy for classification of tomato leaf images into one of the 9 disease categories.  The performance is also evaluated for tomato leaf disease identification from normal leaf images. The process is carried out for 8 different tomato leaf diseases. We have also proposed segmentation-based approach for image classification. Before features extraction, the region of interest is segmented using color-based thresholding method. The proposed model is very useful for providing early advice and warning to farmers. The model can be further extended to operate in real cultivation situation using IOT so that farmers can monitor the farm while sitting at home.

Article Details