摘要
Abstract
To address the problems of complex image backgrounds and a large proportion of small targets in traffic sign images,an improved traffic sign detection algorithm based on MOMP-YOLOv8 is proposed.Firstly,a Mixed Local Channel Attention(MLCA)module is introduced into the neck network,which references spatial and channel attention mechanisms and combines local and global information to improve network detection accuracy and enhance the expressive power of the feature extraction network.Secondly,in the backbone network,Omni Dimensional Dynamic Convolution(ODConv)is used to enrich the capture of contextual information through multi-dimensional attention mechanisms,in order to improve the model's ability to extract target features.Then,a multi-level Adaptive Spatial Feature Fusion(MASFF)detection head is proposed,which adds a 160´160 small object detection head in the detection layer to enhance the detection ability of small objects.Finally,by improving the loss function through the use of PIoU(Powerful Intersection over Union),the convergence speed and detection performance of the model are further enhanced.Through experimental verification on the CCTSDB 2021 traffic sign detection dataset,the results show that the improved algorithm leads the detection accuracy,recall,and mAP values of the baseline YOLOv8n algorithm by 0.9%,1.8%,and 1.4%,respectively.Overall,the detection performance is better than other mainstream object detection algorithms.关键词
交通标志检测/MOMP-YOLOv8/混合局部通道注意力/全维动态卷积/PIoU损失函数Key words
traffic sign detection/MOMP-YOLOv8/mixed local channel attention/omni-dimensional dynamic convolution/PIoU loss function分类
通用工业技术
