Traditional ceramic defect detection mainly relies on manual visual inspection or magnifying glass observation. In order to solve the problems of low detection efficiency and strong subjectivity of results, this paper proposes a ceramic surface defect detection algorithm based on deep learning. According to the specific situation of the defects on the surface of the ceramic cup, a small target detection layer is added on the basis of the YOLOv5 target detection model, at the same time, the position attention mechanism is used for feature reconstruction to improve the detection accuracy, and high-precision defect detection is achieved. According to the actual production of ceramic double-layer cup data acquisition training, and reasoning for each batch of data, the final average detection accuracy reached 95. 4%. The improved YOLOv5 defect detection model in this paper has the advantages of higher accuracy and faster recognition speed, which can greatly reduce the loss of human and material resources and time cost in ceramic quality inspection.