基于注意力机制的改进YOLOv3与交通标志检测

Improved YOLOv3 and Traffic Sign Detection Based on Attention Mechanism

实时而准确的交通标志检测是车辆的辅助驾驶和无人驾驶的关键需求。为解决目标检测算法对小目标物体检测精确率低、检测速度慢的问题,提出一种嵌入混合注意力机制的交通标志检测算法YOLOv3-HA。该算法融合改进的通道注意力机制和子空间注意力机制,使网络模型能够对特征进行通道和空间上的注意力加权,提升网络对有效特征的表达能力并减少干扰特征的影响。采用K-Means++聚类算法对锚框进行聚类和选择,加快网络模型的收敛速度。实验表明,该算法在TT100K(Tsinghua-Tencent 100 K)数据集上的平均准确率均值达到81.0%,相比于YOLOv3算法提升了14.2%;与一些主流目标检测算法相比,YOLOv3-HA算法在准确性和实时性上达到了良好的平衡。

Real-time and accurate traffic sign detection is the key requirement of vehicle assisted driving and unmanned driving.In order to solve the problems of low accuracy and slow speed of target detection algorithms in detecting small target objects,a traffic sign detection algorithm embedded with a hybrid attention mechanism,named YOLOv3-HA,is proposed in this paper.The algorithm integrates the improved channel attention mechanism and subspace attention mechanism,so that the network model can weight the features in channel and space,improve the expression ability of the network to effective features and reduce the influence of interference features.Furthermore,the clustering algorithm K-Means + + is applied to cluster and select the anchor box to speed up the convergence of the YOLOv3-HA algorithm.Experiment results show that the mean average precision of the proposed algorithm on TT100K data set is 81.0%,which is 14.2% higher than that of the YOLOv3 algorithm;Compared with some other target detection algorithms,the YOLOv3-HA algorithm can achieve better accuracy and real-time capability.