添加注意力机制的YOLOv5s算法对带钢表面缺陷检测

Steel surface defect detection based on improved YOLOv5s algorithm

针对传统采用人工检测钢材表面缺陷存在的精度低、速度较慢等问题,将YOLOv5s算法应用在带钢表面缺陷检测的实验中。在原YOLOv5s算法网络上使用新型的C3结构,减小模型的计算量,计算速度有所提升;同时在PANet网络的检测头之前增加NAM注意力机制,提高了网络预测的精度,也加快了整个网络的收敛速度。从试验结果可以得出,神经网络检测的精度和收敛速度都所提升,能够快速准确地定位图片上的缺陷位置。

In view of the low precision and slow speed of the traditional manual detection of steel surface defects,the YOLOv5s algorithm was applied to the experiment of strip steel surface defect detection,the original YOLOv5s algorithm network was modified,and the new C3 structure was used to reduce the calculation amount of the model and improve the calculation speed.The modification on the PANet network,adding the NAM attention mechanism before the detection head,improves the accuracy of network prediction,and also speeds up the convergence speed of the entire network.From the test results,it could be concluded that the accuracy and convergence speed of neural network detection were improved,and it could quickly and accurately locate the defect position in the picture.