Abstract: The traditional manual inspection of river floating garbage is time-consuming and labor-consuming. UAVs and unmanned boats have become the main methods of river inspections. At present， it is limited to manual interpretation of river floating garbage images or simple computer target detection， and there is a lack of automatic classification and detection of floating garbage. In this paper， a dataset of floating garbage in the river channel of the study area is constructed based on drone aerial images， and a deep learning method is used to classify and recognize garbage. Taking into account the unbalanced types of floating garbage in the river and the relatively small proportion in UAV images， the YOLOv5s target detection algorithm has been improved in terms of multi-scale detection and data enhancement. The experimental results show that the improved algorithm improves the mean average precision by 3.47% compared with the original algorithm. The study results show that the combination of deep learning methods and drone technology can efficiently and accurately identify and classify garbage， and provide a basis for decision-making in the treatment of floating garbage in the river.

Key words: UAV, floating garbage, classification detection, deep learning, small goal