In order to correctly extract the lesion of crop disease images, proposes a method to detect color edge of crop disease based on support vector machines and multi feature selection. The method uses (2d + 1) ^* (2d + 1) as the size of the window through the image. It also is used on the image luminance and chrominance channels calculate variance, mean value difference, maximum gradient, characteristics of space position and mean color difference. In order to accurately identify the edge of disease spot, the characteristics of vector-valued input support vector machine is employed. To improve the efficiency of disease spot of edge detection, a method is proposed in the traversal process, if the eigenvalues are smaller than the threshold, skip d row, d column to traverse. The experimental shows that the method has better than the traditional edge detection operator disease spot edge recognition ability.