With the acceleration of the urbanization process and the increase in population density, the amount of garbage generated is also increasing day by day. All kinds of garbage on the road not only affect the appearance of the city, cause various pollutions, but also affect people’s physical and mental health. The existing algorithms for urban road garbage detection have low detection accuracy and may have omissions and false detections for small-scale garbage. In response to this problem, this paper introduces a SCN-YOLOv8 urban road garbage detection algorithm based on YOLOv8. First of all, in order to better capture the detailed feature information of small objects, we propose a Sub-pixel Spatial Attention Mechanism (SSAM). It can enhance the network’s ability to extract detailed information. Secondly, we have designed the Cross-scale Feature Enhance Pyramid Network (CFEPN). CFEPN not only enhances the interaction ability of feature information in shallow and deep networks, but also enables features with rich contextual information to diffuse to various detection scales. Furthermore, in order to improve the stability of model, we adopt the regression loss function based on the Normalized Wasserstein Distance (NWD) metric. Finally, we constructed a Road Garbage Dataset (RGD), which contains ten common types of road garbage. The experimental results show that our model performs better than the baseline under the RGD dataset, with:0.95 increasing by 2.7%, proving that the modifications we made to the original YOLOv8 algorithm are effective.