基于纹理差异度引导的DRLSE病虫害图像精准分割方法

An Accurate Segmentation Approach for Disease and Pest Based on Texture Difference Guided DRLSE

为精准分割纹理背景下的病虫害目标,提出一种基于混合高斯模型的纹理差异度表示方法和基于纹理差异度引导的DRLSE农作物病虫害精准分割方法。采用基于混合高斯模型的纹理差异度表示方法得到纹理差异度图像,利用Otsu自动阈值分割和形态学后处理得到农作物病虫害DRLSE演化的初始区域,将基于改进纹理差异度的边缘停止函数应用到DRLSE演化中。实验结果表明该方法可以获得精准的害虫和病变区域轮廓。

In order to accurately segment diseases and insect pests on crop leaves, a Gaussian mixture model （GMM） based texture difference expression method and an accurate segmentation approach based on improved distance regularized level set evolution （DRLSE） were proposed. Considering that crop leaf generally has a certain texture feature, GMM was used to characterize the leaves texture feature, and improved DRLSE which took full advantage of texture difference information between objects and background, was employed to get the accurate contour of diseases and insect pests. Firstly, the texture feature of leaf sampling area was modeled by using GMM. Secondly, the texture difference between the pixels in the diseases and insect pests＇ area and the sampling area was calculated, and the texture difference image was obtained at the same time. Thirdly, the initial contour for DRLSE was obtained by Otsu and morphologic post processing. Finally, the contour of diseases and insect pests was evolved accurately with texture difference guided DRLSE. The experimental results show that the diseases and pest contour can be obtained accurately with the introduced method and also can provide the basis for the subsequent identification and prevention of crop diseases and pests.