Objective In response to the edge detection of weld seam images in complex environments an improved Canny algorithm for edge detection of weld seam images in complex environments was proposed based on the traditional Canny algorithm replacing Gaussian filtering with median filtering and improving the non-maximum suppression NMS . Methods The weld seam image was subjected to median filtering to preserve edge information and smooth details within the area followed by improved NMS using the local mean deviation of the gradient directions. The influence of gradient amplitude and gradient direction on weld seam edge detection was considered. Finally the dual-threshold method was used to detect and connect edges. The appropriate threshold was selected based on the characteristics of the weld seam image to ensure the accuracy of edge detection results while effectively avoiding some noise interference. Results Compared with the traditional Canny algorithm and other Canny algorithms with improved filters the proposed method achieves the best edge retention and edge linearity. Edge retention can preserve approximately 60% of the original weld seam image and the degree of edge connection is improved compared with the traditional Canny algorithm with enhanced suppression of pseudo edges in the weld seam images. Conclusion The improved Canny algorithm through comprehensive optimization of filtering and NMS algorithms can effectively detect edge information in weld seam images and achieve good edge detection results for weld seam images.