基于改进YOLOv5 的交通标志小目标检测算法

Traffic Sign Small Target Detection Algorithm Based on Improved YOLOv5

针对交通标志小目标和密集目标检测准确率不高的问题,提出了改进YOLOv5s的检测模型。在Backbone网络中添加ECA注意力机制增强小目标交通标志特征信息提取能力;其次采用SPPCSPC结构减少小目标交通标志信息丢失;再使用BiFPN网络融合多尺特征信息,增强融合感知能力;最后将WIoU作为训练时模型的损失函数,降低背景的过度干扰,提升交通标志检测的准确性。试验结果表明,改进后算法的准确率为93.3%、mAP值为92.7%,较未改进前分别提高了2.2%、1.7%。

Aiming at the low detection accuracy of traffic sign small targets and dense targets of traffic signs,an improved YOLOv5s detection model is proposed.It adds ECA attention mechanism to enhance feature information extraction ability of traffic sign small target in Backbone network.Secondly,it adopts SPPCSPC structure to reduce information loss of traffic sign small target.Then,it re-uses BiFPN network to fuse multi-scale feature information to enhance the fusion perception ability.Finally,WIoU is used as the loss function of the model during training to reduce excessive interference of background and improve the accuracy of traffic sign detection.The experimental results show that the accuracy of the improved algorithm is 93.3%,and the mAP value is 92.7%,which is 2.2%and 1.7%higher than before,respectively.