This paper presents a SC-YOLOv7_tiny model based on YOLOv7-tiny, combined with a UAV platform to detect floating garbage in water areas, addressing the low efficiency and high cost of traditional water area floating garbage detection. Firstly, a skip connection-based feature pyramid network (SCFPN) is introduced to detect small objects by combining features of different scales. Secondly, to improve the CIOU loss function, the weight of the angle penalty term θ is used to penalize differences in length and width scales, thus mitigating the problems of feature distribution imbalance, scale mismatch, and difficulty in detecting small objects when combined with the SCFPN feature pyramid. Finally, the SimAM attention mechanism is introduced into the SCFPN feature pyramid module to facilitate the model’s understanding of contextual information and enhance its feature expression capabilities. To verify the effectiveness of the algorithm, this paper uses a public dataset on Universe for testing. Experimental results show that the parameters of the SC-YOLOv7-tiny model are significantly reduced while detection accuracy is improved.