基于显著性轮廓的苹果目标识别方法  

Recognition method of apple target based on significant contour

作　　者：顾苏杭[1,2] 马正华[1] 吕继东[1] 

机构地区：[1]常州大学信息科学与工程学院,江苏常州213164 [2]常州轻工职业技术学院,江苏常州213164

出　　处：《计算机应用研究》2017年第8期2551-2556,共6页Application Research of Computers

基　　金：国家自然科学基金资助项目(51307010);江苏省自然科学青年基金资助项目(BK20140265;BK20140266);常州市科技计划资助项目CJ20160010)

摘　　要：正确地将苹果从图像中识别出来是苹果采摘机器人实现自动采摘的前提,在对该问题研究的基础上,提出一种基于显著性轮廓的苹果目标识别方法。利用K-means无监督聚类算法将图像分割为背景和目标区域;由于光照等因素,目标区域内部存在大面积空洞,引入ASIFT特征,将完整的目标与存在空洞的目标进行特征匹配记录与空洞相对应的特征,由这些特征恢复成像素填补空洞;在基于区域的基础上,采用g Pb轮廓检测器进行轮廓检测,生成较长、较明显的灰度轮廓图像;通过动态阈值Otsu法对灰度轮廓图像进行自动阈值处理,去除目标周围大量的边缘噪声,确定连续的显著性轮廓,完整地提取目标轮廓。实验结果表明,该方法具有更好的准确性与鲁棒性,对苹果目标的正确识别率在98%以上。Correctly identifying apple from image is the premise of apple harvesting robot for achiving automatic picking. On the basis of the study of this problem, this paper proposed a new apple target recognition algorithm based on significant con- tour. Firstly, it used the K-means unsupervised clustering algorithm to segment apple image which was divided into back- ground and apple target area. Due to factors such as light, the apple target region after K-means segmenting contained large hole, so it introduced the ASIFT feature to match between the complete apple target image and the apple target image which in- eluded hole, and recorded the features corresponding to the hole, and filled the hole by pixels relating to those features. Then, based on the area, it detected the apple target image＇ s contour by gPb contour detector to generate longer and moreobvious gray contour image. Finally, it removed much edge noise around the target through the dynamic threshold Otsu for automatic threshold processing to process the gray contour image, and achieved a complete extraction of the contour of the target by deter- mining the continuous and significant contour. Experimental resuhs show that, the proposed approach is more accurate and more robust, the correct recognition rate of apple target is above 98%.

关 键 词：图像分割 苹果 轮廓检测 动态阈值 识别 

分 类 号：TP391.7[自动化与计算机技术—计算机应用技术]