Machine vision based classification of rice cultivars using morphological chromatic and textural features of seed images

Abstract

Variety identification is an important task for plant breeders, farmers and traders. DUS (Distinctness, Uniformity, Stability) protocol is generally carried out for identification of plant variety which is time consuming and laborious. An attempt was made to quantify 28 rice varieties based on seed images by digital image analysis. Rice seed images were captured using Canon-LiDE110 flatbed scanner at 600 dpi resolutions. An algorithm was developed using Matlab 2012B to capture and extract seven morphological, 18 textural features and seven chromatic features. Discriminant analysis was carried out to identify critical parameters and classified them into similar groups. The study identified 14 best features out of 32 features that has capability to discriminate between rice cultivars. Eccentricity, awn length, major axis, equivalent diameter, kernel area, kernel perimeter and minor axis were found to be most critical among morphological features while standard deviation (STD) and Energy were found to be most critical among textural features while Hue, Red and Green were found to be most critical among chromatic features. Thus the present study indicated that morphological, chromatic as well as textural features play a vital role in identification of new varieties and distinguishing them to classify into similar groups.