Abstract: In the production of stainless steel plate, although the traditional contact measurement method can realize the depth detection of steel plate surface defects, it is easy to cause damage due to direct contact with the surface, which affects the measurement accuracy and evaluation reliability. Aiming at the above problems, a method PC-LDA(Point Cloud Layered Depth Analysis) based on point cloud hierarchical analysis is proposed to make up for the shortcomings of existing technology and meet the requirements of accuracy and reliability of surface defect depth detection of stainless steel plate after grinding in industrial applications. In this approach, a blue binocular structured light system is employed to acquire high-resolution point cloud data, which undergoes preprocessing and denoising to improve data quality. The least squares method is then applied to fit a reference plane, calculate the normal vector, and differentiate the sanded regions from the non-sanded surface based on the direction of the normal vector. For the sanded region, depth stratification is performed using distance information from the points to the reference plane. The local peaks and troughs are identified by analyzing the density distribution of the point cloud at each depth level. By calculating the peak widths of local peaks and adjacent troughs and traversing the point cloud data within these intervals using the Euclidean distance, both the hierarchical mean and overall mean depths are obtained, thereby accurately recovering the depth information of the polished region.To validate the accuracy and applicability of the PC-LDA method, experiments were conducted using standard blocks with a depth range of 1.0-1.1 mm and polished stainless steel surfaces from actual production environments. Tests included single-size and multi-size standard blocks. The experimental results demonstrate that the average depth measurement deviation of the PC-LDA algorithm is 0.015 mm, which is significantly below the 0.03 mm threshold. Additionally, the method effectively measures the depth of stainless steel surfaces with high precision, capturing fine geometric features of the surface undulations post-grinding. The proposed PC-LDA algorithm exhibits excellent reliability and consistency, making it well-suited to meet the technical demands of grinding quality assessment and secondary regrinding processes.