基于YOLOv5的小型铝铸件涡轮缺陷检测

Defect Detection of Small Aluminum Casting Turbines Based on YOLOv5

针对小型铝铸件涡轮表面复杂、缺陷过小,难以检测的问题,提出一种改进YOLOv5的小型铝铸件涡轮表面缺陷检测算法。图像预处理采用数据增强策略平衡不同类别样本的分布;使用K-means++算法得到适合本数据集的最佳先验框;对网络的特征提取结构进行改进,在主干网络中添加CA注意力机制模块,帮助模型更加准确的定位和识别;添加小目标检测层,增强对小物体的检测效果。结果表明,改进的算法对小目标缺陷拥有更好的检测效果,平均精度均值(mAP)达到97.8%,满足智能制造自动化生产的需求。

Aiming at the problem of difficulty in detection of small aluminum casting turbine due to the complicated sur⁃face and small defect,an improved YOLOv5 surface defect detection algorithm for small aluminum casting turbine was proposed.The data enhancement strategy was utilized to balance the sample distribution of different categories for image preprocessing.The K-means++algorithm was taken to obtain the optimal prior frame suitable for the data set.The feature extraction structure of the network was modified,and CA attention mechanism module was added to the backbone network,which helps to the accurate positioning and recognition of model.Addition of a small target detec⁃tion layer can enhance the detection effect on small objects.The results indicate that the modified algorithm exhibits a better detection effect on small target defects,where the mean average precision(mAP)reaches 97.8%,meeting the requirement of intelligent manufacturing automation production.