Abstract: Macrostructure inspection of continuously cast slabs is an important procedure for evaluating the internal quality of slabs and guiding the optimization of continuous casting processes. To address the problems of strong subjectivity, low efficiency, and difficulty in data traceability associated with traditional manual visual rating, this study developed a machine-vision-based intelligent inspection platform for slab macrostructures. The platform integrates macrostructure image acquisition, intelligent defect segmentation, automatic rating, and data management. With an improved U-Net model as the core, the platform enables pixel-level recognition of macrostructure defects such as centerline segregation. Based on quantitative indicators including defect area, length, width, equivalent diameter, and the area ratio of the largest connected component, traditional atlas-based rating experience is transformed into computable and traceable digital rating rules. The system establishes a closed-loop workflow covering image acquisition, defect recognition, automatic rating, data traceability, and process feedback. At present, the platform has been implemented at the industrial site of Laiwu Steel Corporation, Shandong Iron and Steel Group, and has achieved stable operation for one year. Field application results show that the usability rate of image data reaches 99.9%, and the overall automatic rating accuracy reaches 98%, demonstrating good engineering application value and promising potential for further promotion.