The YOLOv8s Traffic Sign Detection Method Based on IEMA and Mixed Small Targets

ZHU Qiangjun; HU Bin

doi:10.13501/j.cnki.42-1908/n.2025.06.010

Abstract：Aiming at the problem of low precision in traffic sign detection in complex scenarios,a improved efficient multi-scale attention mechanism and hybrid small target layer-you only look once version 8 small model(AMST-YOLOv8s)was proposed.Firstly,the improved efficient multi-scale attention(IEMA)module was added before the spatial pyramid pooling-fast(SPPF)module in the backbone network to improve the detection precision of traffic signs.Secondly,a hybrid small target detection layer was added to improve the detection precision of small objects in traffic sign detection.Finally,by improving the shape-intersection over union(Shape-IOU),the wise shape-intersection over union(WSIOU)loss function was obtained to replace the complete intersection over union(CIOU)loss function in the model,enhancing the model's ability to detect the diverse shapes of the objects.On the Tsinghua-Tencent 100k(TT100K)dataset,the precision of the AMST-YOLOv8s model was 5.33%higher,the recall rate was 11.30%higher,the mean average precision(mAP)at an intersection over union threshold of 0.5 was 7.78%higher,and the mean average precision(mAP)at an intersection over union threshold from 0.50 to 0.95 was 6.67%higher than those of the original YOLOv8s model.On the Changsha University of Science and Technology-Chinese traffic sign detection benchmark(CCTSDB),stock keeping unit 100k(SKU-100K),visual drones(VisDrone),and visual object classes 2007(VOC2007)datasets,the mAP at an intersection over union threshold of 0.5 of the AMST-YOLOv8s model was 37.10%,0.50%,10.70%,and 1.17%higher respectively than that of the original YOLOv8s model.The results showed that compared with the mainstream traffic sign detection models,the proposed model had the advantages of high precision,strong generalization ability,and light weight.