基于模糊C均值聚类的作物病害叶片图像分割方法研究

Segmentation of crop disease leaf images using fuzzy C-means clustering algorithm

为提高作物病害图像的分割效果,根据作物病害图像的特点,提出了一种基于模糊C均值聚类算法(FCM)的作物病害图像自适应分割方法。该方法将像素的灰度与其邻域均值作为FCM的输入特征,变换FCM的隶属度函数使其包含图像的局部邻域特性;通过聚类有效性验证分析和试验确定模糊C均值聚类算法(FCM)的最优聚类数、模糊加权指数。运用该方法对棉花病害叶片图像进行分割。结果表明:该方法能较好将病斑部分和正常部分分割开,平均分割误差率小于5%,对作物病害图像的分割处理非常有效。

For improvement on segmentation precision of crop disease images, an adaptive segmentation method of crop disease images was proposed based on fuzzy C-mean clustering algorithm （FCM）, according to the properties of crop disease images. The segmentation algorithm used the pixel gray and mean of neighborhood pixels as input features, and modified the membership function of FCM which contains local neighborhood information of image. The optimal cluster number and the degree of fuzziness of FCM were chosen through cluster validity and experiments respectively. The optimal cluster number is 4, and the degree of fuzziness is 2. The adapted segmentation method was used to segment cotton disease leaf images. The result shows that the method of segmentation is satisfactory to separate disease part from normal part of leaves. The mean segmentation errors ≤5%. It is very effective in segmenting and processing crop disease images.