The defects in the metal coating surface of annular ceramic workpiece have significant effects on the conductivity and reliability. Due to the irregularity, small area, and few sample number of defects, it is difficult to achieve efficient and accurate inspection. This paper presents a defect inspection framework based on deep learning for the metal coating surface of annular ceramic workpiece. Firstly, an image acquisition system for the coating surface is designed, and the defects characteristics are analyzed. Then, a surface image data set is constructed through five data augmentation strategies in order to solve the problem of insufficient samples. Finally, a defect detection framework for ceramic metal coating surface based on improved YOLOv7 model is established. By optimizing the clustering algorithm of target box, introducing an attention mechanism, and improving the MPConv structure, the efficient and precise identification of different defects is realized. Experimental results show that the recognition rate of defects including scratch, deficiency, scuffing, and dot is higher than 94%, and the average detection time is about 30 ms. The proposed detection framework based on deep learning shows great application potential in the fields of precise coating and manufacturing of ceramic materials.