基于YOLOv5的表面 缺陷检测优化算法

快速、准确地检测材料表面缺陷已成为各领域研究的重要目标，为增加检测效率，实现设备轻量化，提出了一种基于YOLOv5的目标检测优化算法，添加DyHead检测头，融合多个注意力机制，增强模型的检测精度；更换aLRPLoss损失函数，减少超参数调节工作，优化训练过程；基于FasterNet提出C3-Faster,代替网络中的C3模块，以PConv的思想提升模型检测性能，减少模型体积；最后添加轻量级上采样算子CARAFE,扩大模型感受野，提升对不同大小目标的检测效果。实验结果表明，改进后的YOLOv5模型相比于原版模型，在钢材表面缺陷数据集上总体平均精度提高了4.174%,参数量减少了11.25%,计算复杂度减少了13.75%,权重体积减少了10.72%,检测性能高于SSD、RetinaNet、FCOS、YOLOv3、YOLOv4等主流目标检测算法，在工业检测中具有较高的应用价值。

Optimized Algorithm for Surface Defect Detection Based on YOLOv

The rapid and accurate detection of surface defects in materials has become an important objective across various research domains. To enhance detection efficiency and realize lightweight equipment,this paper proposes a target detection optimization algorithm based on YOLOv5, adding DyHead detection head to enhance the detection accuracy of the model by fusing multiple attention mechanisms; replacing the aLRPLoss loss function to reduce the hyperparameter adjustment work and optimize the training process; propose C3-Faster based on FasterNet to replace the C3 module in the network to improve the model detection performance and reduce the model size with the idea of PConv; finally add the lightweight upsampling operator CARAFE to expand the model perceptual field and improve the detection effect on targets of different sizes. The experimental results show that the improved YOLOv5 model improves the overall average accuracy by 4.174%, reduces the parameter volume by 11.25%, reduces the computational complexity by 13.75%, and reduces the weight volume by 10.72% on the steel surface defect dataset compared with the original model, and the detection performance is also higher than that of SSD, RetinaNet, FCOS, YOLOv3, and YOLOv4 and other mainstream target detection algorithms, which have high application value in industrial detection.