Research on Small Target Detection Technology for River Floating Garbage Based on Improved YOLOv7

Traditional manual inspection of floating garbage in artificial river channels is time-consuming and labor-intensive. Unmanned aerial vehicles (UAVs) and unmanned boats based on computer vision have become the main methods for river channel inspection. In this paper, a dataset of floating garbage in the study area's river channels was constructed based on UAV aerial images. Deep learning methods were employed for garbage classification and recognition. Considering the relatively small proportion of floating garbage in UAV images and the susceptibility to water mist and vapor interference, targeted improvements were made to the YOLOv7 object detection algorithm in terms of multi-scale detection. Experimental results verified that the improved algorithm, compared to the original algorithm, achieved higher detection accuracy for small targets and effectively mitigated water mist interference, with an increase of 2.24% in mean average precision (mAP) for class-balanced accuracy. The study results demonstrate that the integration of deep learning methods and UAV technology enables efficient and accurate identification and classification of garbage, providing decision-making support for the management of floating garbage in river channels.