基于改进YOLO v3的轴承端面缺陷检测算法

为提高轴承端面缺陷检测的速度以及检测精度，提出一种基于改进YOLO v3的轴承端面缺陷检测算法。首先，对图像数据集进行数据增强处理以防止产生过拟合现象；其次，通过改进K-means聚类算法重新聚类出目标检测的Anchor Boxes，并引入SKNet注意力机制模块对原网络结构以及输出层结构进行改进；最后对改进的YOLO v3算法进行实验验证，并与原YOLO v3算法进行对比分析。结果表明，改进后的YOLO v3算法相比原YOLO v3算法对轴承端面缺陷检测的mAP值提升了7.03%，检测速度提升了34.7 帧/s，验证了改进算法的有效性。

Defect Detection Algorithm of Bearing End Face Based on Improved YOLO v3

In order to improve the detection speed and accuracy of bearing end face defects,a defect detection algorithm of bearing end face based on improved YOLO v3 was proposed.The image data set was enhanced to prevent overfitting phenomenon.The improved K-means clustering algorithm was used to re-cluster Anchor Boxes for target detection,and SKNet attention mechanism module was introduced to improve the original network structure and output layer structure.Finally,the improved YOLO v3 algorithm was verified by experiment and compared with the original YOLO v3 algorithm.The results show that the mAP value of the improved YOLO v3 algorithm for the detection of bearing end face defects is increased by 7.03% and the detection speed is increased by 34.7 fps,which verifies the effectiveness of the improved algorithm.