目的 提升自动化产线上工件表面微小缺陷的检测精度和检测速度。方法 首先，在预处理阶段提出采用CutMix的数据增强方法，增加训练样本的多样性，提高模型的鲁棒性和泛化能力，避免训练模型产生过拟合;使用K–means++聚类算法生成边界候选框，以适应不同尺寸的缺陷，并较早地筛选出更精细的特征。其次，借助CSP Darknet53网络及SPP模块提取输入原始图像的特征，通过训练获得针对工件表面质量的在线检测模型，提升YOLOV4缺陷位置检测及识别的精度。结果 实验结果表明，文中所提出的基于YOLOV4的工件表面质量在线监测方法的预测精度达到97.5%，检测速度达到32.8 帧/s，均优于同类的深度学习算法。以贵州某航空工业产品的自动化产线作为实验平台验证了所提方法的可行性和有效性。结论 该方法具备结构简单清晰、自适应性强等优点，检测精度和速度均满足工业场景需求，可以将其用于产品表面质量的在线检测。

The work aims to improve the detection accuracy and detection speed of small defects on surface of workpieces on the automated production line. First of all, the use of CutMix data enhancement method in the preprocessing stage was proposed to increase the diversity of training samples, improve the robustness and generalization ability of the model, and avoid overfitting of the training model. K-means++ clustering algorithm was used to generate boundary candidate boxes to adapt to defects of different sizes and to screen out finer features earlier. Secondly, the CSP Darknet53 network and SPP module were used to extract the features of the input original image, and obtain an online detection model for the surface quality of the workpiece through training, so as to improve the accuracy of YOLOV4 defect location detection and recognition. The experimental results showed that the online monitoring method of workpiece surface quality based on YOLOV4 proposed in this work had a prediction accuracy of 97.5% and a detection speed of 32.8 FPS, which were superior to similar deep learning algorithms. The automated production line of an aviation industrial product in Guizhou was used as an experimental platform to verify the feasibility and effectiveness of the proposed method. Experimental results show that the method has the advantages of simple and clear structure, strong adaptability, etc. The detection accuracy and speed meet the needs of industrial scenarios, and it can be used for online detection of product surface quality.