Abstract: Aimed at the problems of missed detection and false detection in vehicle black smoke detection in traffic scenarios, a vehicle black smoke target detection algorithm based on multi-scale feature fusion is proposed. On the basis of YOLOv5, we reconstructed the multi-scale network, added a small target layer for feature fusion and detection, so that the network’s response ability to small target black smoke was improved. At the same time, when performing multi-scale feature fusion, the path aggregation network (PANet) was replaced by the bidirectional feature pyramid network (BiFPN), which integrated more smoky feature information and adjusted the contribution of smoky features at different scales in the network. The experimental results show that the black smoke detection rate and non-black smoke detection rate of the proposed YOLOv5s-FB method are 92.34% and 91.75%, respectively, and the detection speed can reach 35.3FPS, which can meet the actual application requirements.