融合注意力机制的YOLOv5交通标志目标检测算法

Improved YOLOv5 Traffic Sign Detection Algorithm Based on Fusion Attention Mechanism

为了提高小目标检测的精度和减少漏检率,本文提出了一种基于YOLOv5改进的交通标志目标检测算法。通过融合注意力机制CBMA和加权特征融合网络BiFPN,同时采用SIoU损失函数以提高模型准确性。实验结果表明,本算法在CCTSDB数据集上实现了93.6%的m AP,相比于原始算法提高了6%。尤其在中远距离小目标检测任务下,该算法检测速度达到了63.13FPS,显著改进了原模型效果,满足了实时性要求。

This paper proposes an improved traffic sign object detection algorithm based on YOLOv5,aiming to improve the accuracy of small object detection and reduce the missing detection rate.By introducing the attention mechanism CBMA and the weighted feature fusion network BiFPN,meanwhile,the SIoU loss function is adopted to enhance the model accuracy.Experimental results show that this algorithm achieves a mAP of 93.6%on the CCTSDB dataset,an improvement of 6%compared to the original algorithm.Especially for small object detection tasks at medium to long distances,the algorithm achieves a detection speed of 63.13 FPS,significantly improving the effectiveness of the original model and meeting real-time requirements.