Performance analysis of support vector machine for early identification of citrus diseases

Abstract

Early citrus disease detection is necessary for optimum citrus productivity. But detecting a citrus disease at an early stage requires expert views or laboratory tests. But getting an expert view of all time is impossible for rural farmers. The present study aimed to create a low-cost, intelligent, affordable citrus disease classification system. This study offered a Support Vector Machine (SVM) based smart classification method for categorizing various citrus diseases. Citrus photos were subjected to a variety of image processing techniques to categorize the diseases using SVM and the kernel. Prior to classification, the images were segmented and the hue channel threshold value was used to differentiate the diseased area from the remaining portion of the image. The segmented image’s color and grey domains were used to extract 13 different texture and color features. This study outlined three different SVM kernel types-Linear, Gaussian, and Polynomial, and evaluated their accuracy and confusion matrix performances. The Radial Based Function with a polynomial kernel derived from the SVM outperformed the SVM's linear and Gaussian kernel.