Automatic Orange Fruit Disease Identification Using Visible Range Images

Abstract

China is among the top producers of Navel Oranges in the world. However, the yield of these economically valuable fruits is affected by citrus diseases which cause a reduction in the amount and quality of the produced oranges. Traditionally, farmers rely on human experts to scout the plantations in order to spot infected fruits and identify the diseases. Scouting an entire plantation is a time-consuming task. Also farmers have to incur financial costs in order to pay the experts, who are not always readily available. This study proposes an automatic system for the identification of diseases from images of infected fruits taken by conventional digital cameras. The proposed approach has 5 general steps, background removal using Hough transform for orange shape detection, segmentation of symptoms via thresholding, selection, and extraction of features, and finally, training and classification by majority voting of 3 classifiers, namely, KNN, Random Forest and Multiple Support Vector Machine. The approach is evaluated on 3 diseases of the navel orange fruits namely Citrus canker, Citrus melanose, and Citrus black spot, achieving 93% accuracy using global color histogram, Local Binary Patterns, and Halarick texture features.