Rice variety recognition technique using machine vision

A machine vision system was devised to enable rice variety recognition using a microcomputer. The system is composed of a flatbed scanner with scanning software and at least a 386-based microcomputer with a math coprocessor. This technique provides an objective approach to rice grain variety recognition. A program was created to collect rice grain features such as rice grain area and rice grain perimeter. Feature indexing strategy was applied to four varieties in the IR-line namely IR8, IR42, IR64, IR72. This recognition technique involves sequential comparison of known parameter values to those of the unknown variety. Compactness index derived from area and perimeter indexes was found insufficient as single parameter to recognize rice variety. Histograms generated indicated wide overlaps to effectively use this parameter. The area index also indicative of rice grain profile of the binary image ranged from 2020 to 3784. The perimeter index indicative of the edge detected from binary images ranged from 105 to 165. Compactness index ranged from 4 to 10. Image index was collected and found effective in rice grain variety classification as a single classifier. The image index which is the number of pixels representing the rice grain from the unsegmented image, ranged from 2309 to 4870 for all varieties. Based on the image index the IR-line varieties may be categorized into two groups, Group 1 -comprising IR8 and IR42, and Group 2 - comprising IR64 and IR72. The single classifier was implemented by using IR8 image index values greater than 3600, and IR42 image index values less than 3600 and compactness index values less than 6.5. Similarly, IR72 image index values less than 3100. A double classifier was used to classify IR64 and IR72 rice varieties. It was implemented by using IR64 image index values less than 3600 and compactness index values greater than 6.5. Accuracy of the present system ranged from 81 to 84 per cent for Group 1 using the single classifier, while for Group 2, ranged from 30 to 56 percent, using a double classifier