Abstract: A fast pipeline weld image processing algorithm based on FPGA is proposed for the lack of real-time detection of pipeline weld by serial architecture processor and general computer. In the process of collecting pipeline weld image, there will be noise interference image, which directly affects the edge detection and target recognition of defects. The parallel computing and pipeline technology of FPGA are applied to median filtering and Sobel edge detection algorithm to realize parallel processing between modules, and pixel operation is performed by convolution kernel shift under pipeline coordination. A fast pipeline weld image processing algorithm is realized, which gives full play to the advantages of FPGA computing speed. Experiments show that for a 600 × 400 image, in the environment of a clock frequency of 100 MHz, it only takes 9.56821 ms from pixel reading to image processing completion. Compared with the general computer, the image processing algorithm has a great speed improvement. The algorithm can effectively improve the execution efficiency of image processing and can be used for high real-time pipeline weld image processing.