Abstract: This paper proposed an industrial weld shape detection algorithm for the weld with fish-scale pattern on the surface. A high-dimensional least-square method was used for fitting the weld contour line. The second-derivative of the fitting contour line was calculated to determine the scope of weld toe. Based on the position deviation between fitting and actual contours, the feature points were extracted to obtain the shape parameters of industrial weld. Based on the above algorithm, the hardware and software systems for industrial weld detection were developed and debugged. The test shows that this system is rapid and stable to achieve the acquisition of weld profile data. The accurate results can be obtained through the system to realize the calculation of sampling data.