基于改进YOLOv8的轻量化带钢缺陷检测算法

Lightweight Steel Strip Defect Detection Algorithm Based on Improved YOLOv8

为提升带钢缺陷检测的精准度,并实现模型在移动端的便捷高效部署,提出一种基于改进YOLOv8n的轻量化带钢缺陷检测方法。首先,引入GhostNet来替代网络中的传统卷积层,显著减轻了网络的计算负担,通过加入CA(坐标注意力)关注机制,有效增强了网络的特征提取能力并增强了模型的感受野。其次,在特征融合部分,选用轻量级的CARAFE(基于内容的特征重组)上采样模块,进一步提升了模型对特征的提取效果。最后,为了优化网络边界框回归的性能,采用Wise-IoU边界损失函数来替代原有的损失函数,用改进后的带钢表面缺陷检测方法在NEU-DET数据集上进行实验。结果表明,改进后的方法参数量和计算量分别为9.5 M和6.4 GFLOPs,相较于原始网络提升了15%和21%,同时mAP(平均准确率均值)为81.5%,提升3.2%,优于其他对比目标检测算法,可为移动端检测装备的部署和应用提供参考。

In order to improve the accuracy of steel strip defect detection and realize the convenient and efficient deployment of the model on the mobile terminal,a lightweight steel strip defect detection method is proposed based on improved YOLOv8n.Firstly,GhostNet is introduced to replace the traditional convolutional layers in the network,significantly reducing the computational burden of the network,by adding the CA(coordinate attention)attention mechanism,the feature extraction ability of the network is effectively enhanced and the receptive field of the model is enhanced.Secondly,in the feature fusion part,the lightweight upsampling module CARAFE(content-aware reassembly of features)is selected to further improve the feature extraction effect of the model.Finally,in order to optimize the performance of network bounding box regression,the Wise-IoU boundary loss function is used to replace the original loss function.The improved strip surface defect detection method is used to conduct experiments on NEU-DET data set.The results show that the parameter number and calculation amount of the improved method are 9.5 M and 6.4 GFLOPs,respectively,which are 15%and 21%higher than those of the original network,and the mAP is 81.5%,which is 3.2%higher.It is superior to other compared target detection algorithms,which can provide reference for the deployment and application of mobile terminal detection equipment.