Analysis of Indian Rice Quality Using Multi-class Support Vector Machine

Abstract

AbstractIn this research, quality of Indian rice kernels is accessed using image processing and the techniques of machine learning. Analyzing rice quality through human inspection method is laborious and is prone to error based on subjectivity. Automatic machine vision solution for rice quality analysis is a breakthrough in this research area. The quality parameters considered to classify the rice grains are Degree of Milling (DOM) and Percentage of Broken kernels (PBK). The rice kernels with different DOM are collected from the commercial rice mill. The images of the collected rice samples are captured using an experimental setup. Each image captured has the dimension of 2448 × 3264 pixels. The captured images are pre-processed and features are extracted. The statistical features extracted using GLCM are fed into Machine learning model. The identification of DOM based on these extracted features is carried out using Decision Tree, K-Nearest Neighbor and Support Vector Machine algorithms. This model achieves ≥ 98% of accuracy in identifying PBK. The results obtained shows that SVM with gamma = 0.8 and C = 5 or greater gives a promising results of ≥ 92% for DOM.KeywordsDOMPBKRice kernelsGLCMSVMDecision treeKNN