基于YOLO-GR算法的轻量化钢材表面缺陷检测

Lightweight Steel Surface Defect Detection Based on YOLO-GR

针对工业带钢缺陷检测效率低、精度差、模型部署困难等技术不足,提出了一种改进YOLOv5s算法的轻量化钢材表面缺陷检测模型:YOLO-GR模型。首先,通过引入GhostV2 Bottleneck轻量化模块作为主干特征提取网络,以减少网络的参数量,同时将特征融合网络中的普通卷积块替换为深度可分离卷积块,进一步降低模型的计算复杂度以优化特征提取网络;然后,在检测头部分加入RepLK大卷积核提升网络感受野,以优化大尺度方差的检测效果;最后,引入W-IoU(Wise-IoU Loss)解决了带钢缺陷数据集难易样本不平衡问题,提高模型的泛化性能。实验结果表明,改进后的模型在平均检测精度上比原YOLOv5s模型提升了3.8%,在参数量和计算量比原模型下降了16.6%,模型大小仅仅12 M,为检测模型在移动端上的部署提供了可能。

Aiming at the technical shortcomings of low efficiency,poor accuracy and difficult model deployment of industrial strip steel defect detection,a lightweight steel surface defect detection model based on improved YOLOv5 s algorithm is proposed:YOLO-GR model.Firstly,the GhostV2 Bottleneck lightweight module is introduced as the backbone feature extraction network to reduce the number of parameters of the network.At the same time,the ordinary convolution blocks in the feature fusion network are replaced by deep separable convolution blocks,which further reduces the computational complexity of the model to optimize the feature extraction network.Then,the RepLK large convolution kernel is added to the detection head to improve the network receptive field,so as to optimize the detection effect of large-scale variance.Finally,W-IoU(Wise-IoU Loss)is introduced to solve the imbalance problem of difficult and easy samples in the strip defect data set and improve the generalization performance of the model.The experimental results show that the average detection accuracy of the improved model is 3.8%higher than that of the original YOLOv5s model,and the parameter quantity and calculation amount are 16.6%lower than that of the original model.The model size is only 12 M,which provides the possibility for the deployment of the detection model on the mobile terminal.