Feature Selection by Attribute Clustering of Infected Rice Plant Images

Abstract

Automatic diagnosis of rice plant diseases at an early stage and taking corrective measures in time saves damages of rice crop across the world. The paper aims at developing an appropriate methodology to classify diseases with the help of feature sets obtained by analyzing images of infected rice plants acquired from the field. Since all features are not important in classifying diseases; selection of optimum features is a challenging task to address the problem. The work is performed in three steps. Firstly thirty six features of different category are extracted from the diseased plant images using image processing techniques. Secondly information gain (IG) of each attribute with respect to other attributes is calculated following the concept of information entropy theory. Thirdly using IG, functional dependency of the attributes are evaluated based on which fourteen significant attributes out of thirty six are selected, sufficient to classify the diseases. The proposed method has been applied on four hundred fifty infected rice plant images having three different classes. With the reduced feature set, classification accuracy is calculated using different classifiers demonstrating effectiveness of the proposed model.