水稻種子影像辨識系統之研究

Study on Discrimination System for Seed Characters of Rice with Image Processing

許鐈云(Ciao-Yun Syu) ； 陳宗禮(Chung-Li Chen) ； 黃國益(Kuo-Yi Huang) ； 郭育妏(Yu-Wen Kuo) ； 黃卯昌(Mao-Chang Huang)

《植物種苗》 17卷4期 (2015/12) Pp. 1-18

摘要

水稻為我國最重要之糧食作物，為維持三級良種繁殖制度之水稻種子純度，本研究主要目的為利用影像辨識系統進行水稻種子之品種辨識，針對103年生產之臺南11號、臺稉14號、臺中192號、臺稉9號及臺中秈10號等5個品種的原原種種子，進行種子外觀特徵測量，可有效作為不同品種辨識的外觀特徵包括：長軸、短軸、稃尖底長、稃尖高度、稃尖兩側凹口深度、稃尖尖銳角度、稃尖兩側距離、內外穎寬角度、外穎上邊緣及稃尖顏色等特徵。本系統針對5個水稻品種種子之有效影像擷取率平均為96.9％。系統根據上述的有效特徵進行品種辨識準確率平均達88.8%，其中臺南11號為96.6%，臺稉14號為83.3%，臺中192號為84.1%，臺稉9號為82.6%，臺中秈10號為97.5%。

關鍵字

水稻種子 ； 種子形態 ； 影像處理技術 ； 種子品種鑑定

並列摘要

This study was aim to establish a rapid and accurate discrimination system for rice seeds with image processing in order to maintain the quality and purity of rice seeds under seed propagation system as well as reduce the time and manpower for seed purity analysis. The functions of system included auto-input, counting, capturing image, inspecting rice seed, auto-output, and building image-database. The inspection system focused on foundation seed image-database of 5 rice varieties including Tainan 11, Taikeng 14, Taichung 192, Taikeng 9, and Taichung sen 10 in 2014. To discriminate those varieties, 10 traits including length, width, apiculus length, apiculus height, apiculus lemma depth, apiculus palea depth, apiculus angle, apiculus distance, lemma and palea angle, upper edge of lemma and apiculus color were used in this study. Consequently, the valid image capture rate was about 96.9%, and the average rate for accurate discrimination for variety was about 88.8%. For individual variety, it was 96.6% for Tainan 11 , 83.3% for Taikeng 14, 84.1% for Taichung 192, 82.6% for Taikeng 9, and 97.5% for Taichung sen 10.

並列關鍵字

rice seed ； seed shape ； discrimination system with image processing ； identification of seed variety