We previously showed that a combination of image thresholding, chain coding, elliptic Fourier descriptors, and artificial neural network analysis provided a low false acceptance rate (FAR) and a false rejection rate (FRR) of 11.0% and 19.0%, respectively, in identify Thai jasmine rice from three unwanted rice varieties. In this work, we highlight that only a polynomial function fitting on the determined chain code and the neural network analysis are highly sufficient in obtaining a very low FAR of < 3.0% and a very low 0.3% FRR for the separation of Thai jasmine rice from Chainat 1 (CNT1), Prathumtani 1 (PTT1), and Hom-Pitsanulok (HPSL) rice varieties. With this proposed approach, the analytical time is tremendously suppressed from 4,250 seconds down to 2 seconds, implying extremely high potential in practical deployment.

Date Published: 7 June 2013
PDF: 8 pages
Proc. SPIE 8883, ICPS 2013: International Conference on Photonics Solutions, 888318 (7 June 2013); doi: 10.1117/12.2021861

Published in SPIE Proceedings Vol. 8883:
ICPS 2013: International Conference on Photonics Solutions
Prathan Buranasiri; Sarun Sumriddetchkajorn, Editor(s)