Abstract: During the production process of continuous hot-dip galvanized strip, bright spot defects on the coating surface seriously affect the corrosion resistance of products. In this paper, multi-scale characterization techniques(SEM-EDS and AFM) are used to analyze the morphology and composition of bright spot defects, revealing that their main causes are zinc dross inclusions, gas pore aggregation and abnormal local alloying. An online detection method for bright spot defects based on machine vision is proposed, and an improved U-Net network is adopted to realize real-time classification of bright spot defects, finally achieving a classification accuracy of 98.2% and above, which meets the requirements of real-time performance and high precision for bright spot defect detection in high-speed production lines of continuous hot-dip galvanized strip. The correlation between coating thickness control precision and bright spot defect suppression is studied, a fuzzy PID controller integrated with dynamic weight adjustment is designed, and the feedback data of laser thickness gauge is incorporated to realize adaptive control of coating thickness(control error within ±1.5 μm). Experiments show that this scheme reduces the defect occurrence rate by 63% and increases the coating thickness qualification rate to 99.1%.