基于RGB植被指数的大田油菜图像分割定量评价
Quantitative evaluation of in field rapeseed image segmentation based on RGB vegetation indices

以自然光下大田油菜幼苗图像为研究对象，运用超红指数ExR、超绿指数ExG、超绿超红差分指数ExGR、归一化植被指数NDI、植被提取颜色指数CIVE、植被指数组合COM 等6种常用植被指数和阈值算法分割具有阴影区域的大田油菜图像，同时试验中引入定量评价标准客观评价常用RGB空间植被指数的分割效果。结果表明：定性分析中COM指数优于其他5种植被指数，能够减少阴影带来的分割影响，并在局部叶片分割试验中保留完整叶片轮廓；定量分析中COM指数提供最佳分割精度、灵敏度和特指度分别为94.1%、97.2%、90.9%，其相应标准差为1.1、1.3和0.06。
In field rapeseed seedling images under natural illumination were studied with six color vegetation index segmentation approaches including excess red vegetative index (ExR),excess green vegetation index (ExG),excess green minus excess red (ExGR),normalized difference vegetation index (NDI),color index of vegetation extraction (CIVE) and combination of vegetation indices (COM).The thresholding based algorithms were used to extract in field rapeseed plant with shadow image.The segmentation of common RGB vegetation indices were objectively estimated with quantitative evaluation criteria.The results showed that the COM index is superior to the other 5 vegetation indices in the qualitative analysis,which can reduce the segmentation effect caused by the shadow and retain the complete blade profile in the local blade segmentation tests.In the quantitative analyses,the COM index provides the best segmentation accuracy,sensitivity and specificity of 94.1%,97.2% and 90.9%,respectively,with the corresponding standard deviations of 1.1,1.3 and 0.06