Aiming at the shortcomings of Canny's algorithm in the selection of high-density salt-pepper noise and double threshold, an improved method is proposed, in which multi-scale adaptive anisotropic diffusion equations are used to smooth images in order to effectively remove the high-density pulse signal and better retain edge profile information. In the selection of double threshold, the OTSU method is used to select an adaptive threshold according to the gray information of images, which can better meet the real-time requirements. To verify the algorithm's performance, experiments were carried out under different noise concentrations. Experimental results show that the improved edge detection algorithm can keep a complete set of edges, improve the accuracy of edge detection and get ideal edge profile information.