Detection and Classification of Citrus Leaf Disease Using Hybrid Features

Abstract

Agriculture has played important role in the rise of human civilization. It is the backbone of our economic system and is highly dependent on the horticulture. Various diseases may affect the plants which are to be handled by the farmers within time to increase their productivity. Most diseases affecting plants can be diagnosed at an early stage by using their leaf to improve both quality and quantity of fruits. But detection of leaf disease at an early stage is a challenging task. So, to overcome this situation various researchers have presented different techniques some of which were very expensive and can be used by only trained persons. This paper presents a technique for detection and classification of citrus leaf diseases based on texture and color features extracted after using CES enhancement and segmentation of the diseased part. Segmentation is done by using k-means clustering technique. Color features are extracted by separating each component in HSV, YCbCr and LAB color spaces. Feature selection has been done based on ANOVA F-test to skip irrelevant features. Finally, the classification is done with support vector machine, linear discriminant analysis, k-nearest neighbors and multi-layer perceptron. The accuracy parameter is used to analyze the performance of proposed method. The results obtained are quite encouraging.