Indian Journal of Science and Technology

Abstract

Background: Rice (Oryzasativa) is one of the most widely consumed staple foods, especially in the Asian subcontinent. Traditionally, grain testers who make use of calipers and other specialized tools for measuring features of grains are employed for classifying rice grains into different varieties. However, besides size and shape, various other characteristics, such as whiteness and chalkiness, also contribute towards its taste and overall quality. Objective: Computerized classification of rice grains can help in reducing errors of manual grading and in considering more features that indicate quality. The proposed work considers few commercially available rice grains in the South-Indian region to identify new attributes for better grain classification. Methodology: Rice samples were collected from rice outlets across Karnataka. Images of the grains were captured using flat-bed scanning technique and processed to extract new features that represented chalkiness and whiteness of grains along with other morphological features such as area, perimeter, major and minor axes, etc. Machine learning algorithms were used to create classification rule. Findings: The new features extracted were found to contribute significantly towards the classification. Nine varieties of rice grains were considered for the study and the system was able to successfully classify the grains with an accuracy of 95.78% using the NB Tree and SMO classifiers. Novelty: Many studies that consider the morphological features of grains such as its area, shape etc. have already been performed. However, the shapes and sizes of the different varieties are too varied to generalize a common formula for the classification of all varieties of rice. In this paper, Fourier features are also extracted from grain images in addition to the spatial features to arrive at an improved accuracy for classification. 

Keywords: Fast Fourier Transform (FFT), Grain Classification, Naive Bayes Tree (NB Tree), Rice (Oryzasativa), SMO (Sequential Minimal Optimization)