Abstract: During the use of heat transfer labels, due to changes in the temperature of the heat transfer drum, problems such as wrinkles and incomplete demolding of the heat transfer labels may occur during the transfer process. In the problem of defect detection in heat transfer printing, some larger labels have many types of defects and some unknown defects. Therefore, a multi-class label defect detection method based on YOLO network model was proposed. The self-adaptive matching defect detection method was combined with the improved YOLO network model for detection, and the YOLO network model was added with an attention mechanism to improve the model's ability to detect small target defects. During the processing, first, the labels in different regions were quickly located for preprocessing; Then, different regions were detected using different detection methods; Finally, the results of different regions were fused to determine the detection results. The results showed that the multi-class label defect detection method based on the YOLO network model can effectively detect defects in heat transfer labels, with an overall detection accuracy of 98%, which can meet practical production requirements.