自然场景下树上柑橘实时识别技术  

Real-time Recognition of Citrus on Trees in Natural Scene

作　　者：吕强[1] 蔡健荣[1] 赵杰文[1] 王锋[1] 汤明杰[1] 

机构地区：[1]江苏大学食品与生物工程学院,镇江212013

出　　处：《农业机械学报》2010年第2期185-188,170,共5页Transactions of the Chinese Society for Agricultural Machinery

基　　金：国家自然科学基金资助项目(30771243);国家"863"高技术研究发展计划资助项目(2006AA10Z263)

摘　　要：采用G-B色差分量,通过Otsu自适应阈值分割算法分割成熟柑橘图像,采用基于距离变换的分水岭分割将遮挡、重叠柑橘逐个分开,利用凸包算法修复遮挡区域,设定圆度阈值去除误分割区域,然后提取柑橘轮廓,并运用基于Tukey权重函数的最小二乘圆拟合特征圆,提取了柑橘的中心坐标及半径。对87幅图像中592个柑橘进行了识别试验,总体识别率达87.2%,遮挡或重叠柑橘识别率均超过80%。Using the gray image of color difference G-B,Otsu algorithm was employed to segment the images of mature citrus and background.Then the occluded fruits and overlapped fruits were separated by watersheds method with distance transform.Convex hull algorithm was used to repair the occluded regions.Roundness threshold was set to remove the error segmented regions.After the citrus contours were extracted,the contours was fitted by means of the least-squares circle fit with Tukey weight function,and the centroid coordinates and radius were gained.The recognition results of 592 fruits in 87 images showed that the total correct recognition rate is up to 87.2%,and the correct recognition rate of occluded citrus and overlapped citrus are more than 80%.

关 键 词：柑橘 图像识别 分水岭分割 最小二乘圆拟合 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] S666.2[自动化与计算机技术—计算机科学与技术]