基于FMS-YOLOv5s的轻量化交通标志识别算法

Lightweight traffic sign recognition algorithm based on FMS-YOLOv5s

针对目前的道路交通标志模型有着检测速度慢、模型大和参数多的缺点,提出了一种基于YOLOv5s算法的轻量化交通标志识别算法。首先引入轻量化FasterNet网络,利用该网络中的FasterNet Block结构与原主干网络的C3融合,形成一种全新的C3Faster结构;接着将原网络的损失函数修改为基于最小点距离(MPDIoU)的损失函数,来提高边界框回归的准确性和效率;最后结合高效且轻量的置换注意力机制(shuffle attention,SA),提高模型的泛化能力和稳定性。在CCTSDB 2021数据集上的实验结果表明,与原网络相比,改进后模型的参数量、模型大小、GFLOPs分别减少了17.5%、17.5%和20%;同时mAP@0.5、mAP@0.75、mAP@0.5:0.95分别提升了2.3%、3.4%和2.4%。而且与YOLOv3-tiny等其他算法对比,所提出的算法有明显的优越性,能满足各种场景下移动端实时性的需求。

A lightweight traffic sign recognition algorithm based on YOLOv5s algorithm is proposed for the current road traffic sign model with the disadvantages of slow detection speed,large model and many parameters.Firstly,a lightweight FasterNet network is introduced,and the FasterNet Block structure in the network is fused with the C3 of the original backbone network to form a new C3Faster structure.Then the loss function of the original network is modified to MPDIoU to improve the accuracy and efficiency of the bounding box regression.Finally,the efficient and lightweight SA attention mechanism is combined to improve the generalization ability and stability of the model.The experimental results on the CCTSDB 2021 dataset show that compared with the original network,the number of parameters,model size,and GFLOPs of the improved model have been reduced by 17.5%,17.5%,and 20%,respectively.Meanwhile,mAP@0.5,mAP@0.75,and mAP@0.5:0.95 have been improved by 2.3%,3.4%,and 2.4%,respectively.And comparing with other algorithms such as YOLOv3-tiny,the proposed algorithm has obvious superiority and can meet the real-time demand of mobile in various scenarios.