复杂背景下黄瓜病害叶片的分割方法研究  

Segmentation method for cucumber disease leaf images under complex background

作　　者：张芳[1] 王璐[2] 付立思[1] 田有文[1] 

机构地区：[1]沈阳农业大学信息与电气工程学院,辽宁沈阳110161 [2]东北大学信息科学与技术学院,辽宁沈阳110819

出　　处：《浙江农业学报》2014年第5期1346-1355,共10页Acta Agriculturae Zhejiangensis

基　　金：国家自然科学基金(61202258);辽宁省科学事业公益研究基金(2012005015);辽宁省博士后集聚工程项目(2011921012)

摘　　要：利用图像处理和模式识别技术进行复杂背景下黄瓜叶部病害的自动识别,需要先把目标叶片从复杂背景中分割出来,才能进行后续的特征提取和病害识别。为实现复杂背景下黄瓜叶片的分割,首先利用K-均值聚类算法去除图片中的非绿色部分,再采用基于laplacian of gaussia(LOG)算子的方法对待分割的叶片进行区域检测,然后进行基于形状上下文(shape context)的模板匹配和分割。为了提高匹配速度,先检测叶片的生长点和叶尖,以确定叶片的位置、尺寸和方向;然后使用基于超像素(superpixel)的最优匹配搜索方法来减少搜索的复杂度。对20幅黄瓜叶部病害图像进行分割测试,并与人工分割法进行对比,结果表明,本文所采用的分割算法能较好地从复杂背景下提取出黄瓜叶部病害图像,分割准确率达94.7%,为后期黄瓜病斑的特征提取等工作奠定了良好的基础。In order to realize automatic identification of cucumber disease leaves in the complex background,target leaves should be segmented from the complex background first to facilitate the subsequent feature extraction and disease recognition. For this purpose,K-means clustering algorithm was initially used to remove the non-green parts of the image,and then the approach based on LOG operator was proposed to select the candidate leaf areas. Finally,template matching was conducted based on shape context. During the matching process,the position,size and direction of the leaves were firstly identified via the detection of the growing point and apex of leaves to improve the matching efficiency,along with the search for the optimal matching based on superpixel to reduce the search complexity.To evaluate the feasibility of the proposed segmentation approach,20 images of cucumber diseased leaves were segmented,and the result was compared with manual segmentation. It was shown that the proposed segmentation approach could extract images with cucumber diseased leaves from the complex background,and the average segmentation accuracy rate was 94. 7%,which built a solid foundation for the subsequent feature extraction of cucumber lesion.

关 键 词：图像分割 K-均值聚类 模板匹配 形状上下文 黄瓜叶片 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置] TP391[自动化与计算机技术—控制科学与工程]