基于机器视觉的大豆细菌斑点病粒检测  

Inspection of soybean with spots caused by bacteria based on machine vision

作　　者：时玉强[1] 柴玉华[1] 郑先哲[1] 

机构地区：[1]东北农业大学工程学院,哈尔滨150030

出　　处：《东北农业大学学报》2009年第4期106-110,共5页Journal of Northeast Agricultural University

基　　金：中国博士后基金(20070410833);黑龙江省博士后资助项目(LBHZ_07233);东北农业大学博士启动基金资助项目

摘　　要：文章基于机器视觉,通过图像获取系统得到大豆的表面颜色特征,应用SAS对大豆表面颜色特征进行LOGISTIC回归后,应用BP神经网络对大豆进行标准粒与细菌斑点病粒的分类。经过网络训练后,选用收敛效果好的网络对数据进行仿真预测,共计160粒,其中标准大豆80粒,细菌斑点病80粒。得到的测试识别率为:标准大豆96.3%、大豆菌斑粒98.8%。本研究为大豆菌斑粒的在线识别提供了一定的依据,有利于实现大豆的在线缺陷粒检测。Machine vision has been widely used on detecting and determining defection. Spots on soybean caused by bacteria were inspected with this technology. Nine soybean color properties （R, G, B, H, S, V, G/B, G/ R, R/B） were extracted by computer from digital images. After analyzing these traits by logistic regression with statistics analysis system, BP Neural Network was used to distinguish normal soybean seeds and spotted ones,160 ones in total. The rate for normal soybeans was 96.3%, and for spotted ones was 98.8%. The research results provide a foundation for soybean online identification.

关 键 词：颜色特征 机器视觉 BP神经网络 

分 类 号：S565.1[农业科学—作物学]