Abstract: The texture characteristics of defects in the welding area are slightly different from those in the normal area, and are easily affected by background noise interference, making it difficult to accurately distinguish and resulting in a decrease in the accuracy of defect detection. Therefore, the paper proposes a non-destructive testing method for welding zone defects in orthotropic steel structure bridge deck panels. Using magneto-optical imaging equipment to capture images of the welding area, using wavelet decomposition for multi-scale sub-band analysis of the images, while covering details and contour information, effectively distinguishing differences in texture features. According to the symbol pattern partitioning method, the sub-band coefficients are converted into symbol sequences to enhance the robust expression of local textures. The sampling direction mean vector is used to analyze the statistical characteristics of textures in different directions, suppress random noise, highlight the directional features of defects, and achieve deep mining of multi-scale texture features in the welding area, effectively avoiding the influence of noise interference; Integrate the obtained features into a vector set and input it into the K-means clustering algorithm. By calculating the distance and iteratively updating the clustering center, the defect area is divided and the type is preliminarily determined; Based on the initial positioning of defects, the Hermite interpolation method is used to generate edge curves of the defect area based on key points and their related information, and finally complete non-destructive testing to ensure accurate restoration of the true contour of the defect. The experimental results show that the proposed method can accurately determine the location of defects in the panel welding area with high feature coverage.