改进K⁃means算法的玉米叶部病害图像分割研究

Maize leaf disease image segmentation based on improved K⁃means algorithm

为了提高玉米叶部病害图像分割的准确率,提出一种基于Lab颜色空间的改进K⁃means聚类算法。针对叶部病害图像中病斑区域与正常区域存在的颜色差异性,选择在Lab颜色空间进行分割处理。采用K⁃means聚类算法在分割过程中会存在初始聚类中心点难以确定、分割时间较长、边缘信息分割不完善等问题,文中通过两方面对K⁃means聚类算法进行改进:首先是在Lab颜色空间中的a,b两个颜色通道搜寻波峰,确定初始聚类中心点的位置和数量;其次是用马氏距离替换欧氏距离进行距离度量的优化。应用改进后的K⁃means聚类算法对60幅玉米病害图像进行分割,平均误分率为5.72%,平均分割时间为6.69 s,与传统的分割方法相比,分割准确率提升,分割时间缩短。实验结果表明,改进K⁃means算法能实现玉米叶部病害图像的快速、准确分割。

An improved K⁃means clustering algorithm based on Lab color space is proposed to improve the accuracy of maize leaf disease image segmentation.In allusion to the color difference between the diseased area and the normal area in the maize leaf disease image,the segmentation process was carried out in Lab color space.By means of the K⁃means clustering algorithm,there are some problems in the segmentation process,such as difficult to determine the initial clustering center point,long segmentation time,imperfect edge information segmentation and so on.The K⁃means clustering algorithm was improved in two aspects,the one is that the wave crests in a and b color channels in the Lab color space were searched to determine the location and number of initial clustering centers,another is that Mahalanobis distance was used to replace Euclidean distance to optimize the distance measurement.The improved K⁃means clustering algorithm was used to segment 60 maize leaf disease images,the average segmentation error rate was 5.72%,and the average segmentation time was 6.69 s.In comparison with the traditional segmentation method,the segmentation accuracy was improved and the segmentation time was shortened.The experimental results show that the improved K⁃means algorithm can segment maize leaf disease images quickly and accurately.