Design of a classifier for tomato leaf disease identification

Abstract

Plants are the backbone of human existence for they are directly depended on for food. Plant infections and diseases are thus a major concern. Technology can promote food production in several ways through the application of computer vision technology that employs image processing to determine several aspects. Faster and timely plant disease recognition could immensely aid in the early application of appropriate treatment methods that fundamentally reduce economic losses. The introduction of machine learning techniques in image classification has revolutionized digital imaging and learning systems. Presently, convolutional neural networks have been found to provide the most accurate results while grey level cooccurrence is a popularly used descriptor. However, Convolution Neural Network (CNN) requires numerous learning iterations which lead to high computation costs whereas Grey Level Co-Occurrence Matrix (GLCM) cannot be used alone as a descriptor because a classifier is required to carry out the classification of the texture features extracted. This study proposed a hybrid model that combines CNN and GLCM techniques to classify plant diseases from a set of plant images. The research methodology used a systematic literature review and experimental research design. The systematic literature review was employed to determine and identify the existing techniques in digital plant images and the features to be used in classifying plant diseases. In experimentation, the study evaluated GLCM contrast, energy, and correlation features. The classification was carried out in three phases; 100, 150, and 200 iterations where the GLCM-CNN network had the best accuracy of 96.09% and F1 score of 0.8884 using energy texture images with 200 iterations.