基于机器视觉的苹果识别和形状特征提取  

Recognition and Shape Features Extraction of Apples Based on Machine Vision

作　　者：司永胜[1,2] 乔军[3] 刘刚[1] 刘兆祥[1] 高瑞[1] 

机构地区：[1]中国农业大学现代精细农业系统集成研究教育部重点实验室,北京100083 [2]河北农业大学信息科学与技术学院,保定071001 [3]中国农业大学网络中心,北京100083

出　　处：《农业机械学报》2009年第8期161-165,73,共6页Transactions of the Chinese Society for Agricultural Machinery

基　　金：国家"863"高技术研究发展计划资助项目(2006AA10Z255)

摘　　要：提出了利用色差R-G和色差比(R-G)/(G-B)相结合的苹果识别方法。在顺光、逆光等不同情况下对拍摄的苹果图像进行了识别,并对识别后的图像进行消除噪声、区域填充等预处理,获得苹果的轮廓图像。针对轮廓图像,采用遗传算法进行形状特征提取。采取多次运行遗传算法,并依次转换目标轮廓点为背景点的方法,处理果实图像邻接、重叠问题。实验结果表明:苹果识别方法在一定程度上消除了阴影、逆光、土壤等影响,识别率达97%。基于遗传算法的形状特征提取方法,可对邻接、重叠图像进行有效分割,快速、准确地实现苹果图像圆心坐标和半径的提取。An apple recognition method with color difference R - G and color difference ratio （ R - G ）/（ G - B ） was presented. The apples in the images taken under frontlighting and backlighting conditions were recognized. The contour images were gained through the pretreatments such as noise removal, area filling etc. Shape features were extracted from the contour images based on genetic algorithm （GA）. For the clustered or overlapped apples, the method of multiple running GA and converting the contour points of every apple into background in turn was applied. The experiment results showed that the proposed recognition method, to a certain extent, eliminated the influence of shade, backlighting and soil. The recognition rate reached 97 %. The features extraction method based on GA segmented the clustered or overlapped apples successfully. Meanwhile, the center points and radiuses were quickly and accurately extracted.

关 键 词：苹果 机器视觉 图像识别 遗传算法 特征提取 

分 类 号：TP242.62[自动化与计算机技术—检测技术与自动化装置] TP391.41[自动化与计算机技术—控制科学与工程]