基于改进YOLOv7的小目标交通标志检测算法

Small Target Traffic Sign Detection Algorithm Based on Improved YOLOv7

在自动驾驶中,检测小目标的精度和漏检问题对于车辆设备对路况判断的准确性至关重要。针对小交通标志检测精度漏检的问题,文中提出一种基于YOLOv7的小目标交通标志检测算法。在Neck中引入ACmix注意力模块,利用ACmix注意力模块提高网络对小尺度目标的敏感度,降低噪声所带来的影响;在输出端引入多头自注意力机制SPD卷积构建块,提升输出端对交通标志小目标的检测性能;使用SIoU替换原始YOLOv7网络中的CIoU来优化损失函数,提高网络鲁棒性。为验证所提算法的性能,在CCTSDB交通标志数据集上进行了实验验证。实验结果表明,对于数据集中的小目标,改进后的YOLOv7网络相比于原始的YOLOv7网络,漏检情况得到了明显改善,且map达到了98.7%,运行速度达到了110.5 fps。相较于原始YOLOv7模型,精度和速度均有提升,可以满足交通标志检测的要求。

In automatic driving,the problem of detecting the accuracy and leakage of small targets is crucial for the accuracy of vehicle equipment in judging road conditions.Aiming at the problem of small traffic sign detection accuracy and missed detection,a small target traffic sign detection algorithm based on improved YOLOv7 is proposed.The ACmix attention module is introduced into Neck,which is used to improve the sensitivity of the network to small-scale targets and reduce the impact of noise;the multi-head self-attention mechanism SPD convolutional building block is introduced at the output side to improve the detection performance of small traffic sign targets at the output side;and the loss function is optimized by using SIoU to replace the CIoU in the original YOLOv7 network that improves the robustness of the network.To verify the performance of the proposed algorithm,experimental validation is carried out on the CCTSDB traffic sign dataset.The experimental results show that for the small targets in the dataset,the improved YOLOv7 network,compared with the original YOLOv7 network,the leakage detection is significantly improved and the map reaches 98.7%,and the operation speed reaches 110.5 f/s,which is an improvement in both accuracy and speed compared with the original YOLOv7 model,and it can satisfy the requirements of traffic sign detection.