基于麦穗特征小麦品种BP分类器设计.PDF

中国农学通报 2011,27(6):464-468 Chinese Agricultural Science Bulletin 基于麦穗特征的小麦品种BP分类器设计 毕 昆，姜 盼，唐崇伟，黄菲菲，王 成 （北京农业信息技术研究中心，北京100097） 摘 要：小麦的穗部形态参数是直接反映小麦生长状况的重要参数,也是育种和考种专家关心的重要参 数。但由于小麦的品种繁多，发展新型技术快速识别小麦品种是减少工作人员工作量的有效措施。基 于数字图像分析，利用小麦穗部芒个数、芒长、穗长、RGB 颜色的外部形态特征，对新疆的4 个春小麦品 种共40 个样本进行了分类识别。利用提取的其中8 个特征参数，设计了一个3 层的BP 神经网络分类 器，平均准确识别率在85% 以上，其中2 个小麦品种的准确识别率达到了100% 。通过图像处理的方法 提取小麦的芒长、芒个数和穗长并基于这些参数来识别小麦的品种完全可行。 关键词：小麦识别；BP 神经网络；颜色特征；芒个数；穗长 中图分类号：TP391.4 文献标志码：A 论文编号：2010-2527 TheDesignofWheatVarietyBPClassifierBasedOnWheatEarFeature Bi Kun, Jiang Pan, Tang Chongwei, Huang Feifei, Wang Cheng （BeijingResearchCenterforInformationTechnologyInAgriculture, Beijing 100097） Abstract: Morphological parameters of wheat spike are direct reflection of the wheat growth status and the most concerned parameters of the breeding experts and testing experts. The development of wheat variety rapid identification technology is a necessary measurement to reduce workload and increase efficiency. Digital image analysis was adopted in the identification of 40 samples of Xinjiang spring wheat of four varieties according to the spike external morphological characteristics: awn number, awn long, ear length, and RGB color features. A three layer BP neural network classifiers was designed through the eight extracted characteristic parameters. The average accurate recognition rate was more than 85%, among which, there were two varieties ’accuracy rate reaching 100% , thus proved that extracting wheat mount long, mountain number and panicle length parameters by image processing to identify the variety of wheat was feasible. Keywords:wheat recognition; BP neural network; color feat