基于YOLOv5s网络改进的钢铁表面缺陷检测算法

Improved Steel Surface Defect Detection Algorithm Based on YOLOv5s Network

针对目前钢铁表面缺陷检测算法存在检测精度低、检测速度慢和模型复杂度高等问题,提出基于YOLOv5s改进的钢铁表面缺陷检测算法。将SE通道注意力模块融入骨干网络中以增大缺陷特征通道权重,降低背景干扰,提高算法对缺陷特征的提取能力;在颈部网络融入STR多头自注意力模块,提高缺陷边缘纹理等细节特征的比重;改进损失函数为SIoU,缩短预测框回归收敛过程以提高算法检测速度。实验结果表明:改进算法在NEU-DET数据集上的mAP值为80.4%,较YOLOv5s提高5.5%,每秒处理帧数为100,算法体积降低约8.3%,算法计算量降低约4.3%,对比其他的目标检测算法,改进算法在检测精度、检测速度上均明显提升,模型复杂度降低明显。改进算法可满足实时钢铁表面缺陷检测需求。

Aiming at the problems of low detection accuracy,slow detection speed and high model complexity of the current steel surface defect detection algorithm,an improved steel surface defect detection algorithm based on YOLOv5s was proposed.The SE channel attention module was integrated into the backbone network to increase the weight of defect feature channels,reduce background interference,and improve the extraction ability of the algorithm for defect features.The STR multi-head self-attention module was integrated into the neck network to increase the proportion of detail features such as defect edge texture.The loss function was improved to SoU to shorten the prediction box regression convergence process and improve the algorithm detection speed.The experimental results show that the mAP value of the improved algorithm on the NEU-DET dataset is 80.4%,which is 5.5%higher than that of YOLOv5s,the number of processed frames per second is 100,the algorithm volume is reduced by about 8.3%,and the algorithm calculation amount is reduced by about 4.3%.Compared with other target detection algorithms,the improved algorithm has significantly improved the detection accuracy and detection speed,and the complexity of the model is significantly reduced.The improved algorithms meet the needs of real-time steel surface defect detection.