With the development of passenger vehicles, high-speed trains, and other transport vehicles toward higher speeds, lighter weight, and longer service life, coating surfaces serve not only as important exterior finishing layers but also as critical protective barriers against corrosion, oxidation, and environmental degradation. During manufacturing, microscopic defects such as scratches, pinholes, and particles are inevitably generated on coating surfaces. These defects can compromise coating integrity, accelerate corrosion of the underlying substrate, and, in severe cases, pose risks to operational safety. Therefore, highly reliable, efficient, and precise detection of coating surface defects is of critical importance. However, the highly reflective nature of coating surfaces, together with the complex morphology and heterogeneous characteristics of various defects, poses significant challenges to conventional vision-based inspection, resulting in fundamental limitations in “seeing clearly,” “detecting accurately,” and “measuring precisely.” To address these challenges, the research team first developed a defect-enhanced imaging method based on modulated intensity decoding (MID). Through active optical modulation at the physical imaging level, surface reflections are transformed from interference into an information carrier that can be encoded and decoded, enabling effective enhancement and stable acquisition of defect images. Second, a deep learning model for defect detection was developed based on hierarchical fusion of discriminative features, addressing the mutual interference caused by highly coupled and confusing information and enabling accurate identification of multiple types of complex defects. Furthermore, a prior-constrained weakly supervised defect segmentation method was proposed. Using only limited bounding-box-level annotations, the method achieves pixel-level defect segmentation and enables accurate measurement of defect dimensions. Finally, optical imaging, machine vision, intelligent recognition, robotic motion control, and industrial communication technologies were integrated to develop an automated vision inspection system for vehicle coating surface defects, enabling automatic image acquisition, accurate defect identification, localization, and statistical analysis. The developed technologies have been successfully applied in industrial scenarios such as automotive manufacturing, providing technical support for the automated and intelligent inspection of highly reflective vehicle surfaces. They also offer an extensible technical pathway for visual inspection of other highly reflective industrial surfaces, including high-speed train bodies, silicon wafers, and smartphone cover glass.
 

defect detection,transport vehicles,highly reflective imaging,multi‑scale feature fusion,weakly supervised segmentation