Milled Rice Grain Grading using Raspberry Pi with Image Processing and Support Vector Machines with Adaptive Boosting

Abstract

Rice is a staple food in many countries. The price of rice depends on the qualities that are often quantified based on color, size, and presence of some regional color information. In the Philippines, the National Food Authority released the National Grain Standards for milled rice grains to facilitate the uniform classification of rice. The standards specify the grades: Premium and Grade 1–5 to grade milled rice grain samples based on the number of immature, red, fermented, chalky grains, and others, present in the sample. This study aimed to design and develop a standalone system capable of grading rice samples using grain validation, color and area analysis, and support vector machines with adaptive boosting. The image acquisition platform was created to provide a constant lighting setting and an enclosed staging platform capable of extracting an average of fifty grain images per sample. Seven support vector machine classifiers boosted with adaptive boosting, one chalky classifier, one grain size classifier, were created, trained, and tested. Feature vectors for the SVMs were histogram of gradients features and the color histogram properties: mean, skew, and dominant. The evaluation of the device resulted with an overall micro-average precision of 0.8667 and a micro-average recall of 0.8667 with an Fl-Score of 0.8667.