Low-Altitude Economy Meets Safety Management: Leveraging a Multi-Attention High-Resolution Network for Enhanced UAV Detection

In recent years, low-altitude, slow-speed, and small-sized multi-rotor UAVs (Unmanned Aerial Vehicle) have been widely used in all aspects of daily life. While the rapid development of the UAV industry has introduced numerous conveniences into our daily lives, the misuse and improper operation of UAVs have imposed higher demands on regulatory frameworks and presented new challenges to public safety management. The irregular flight of multi-rotor UAV has become an important factor threatening national security, airspace flight safety, and public safety. At present, the use of optoelectronic imaging systems to identify multi-rotor UAV is a common method. In this paper, inspired by the Area Attention mechanism, channel attention mechanism and small-size target detection design, an improved model based on the YOLO framework is proposed. Furthermore, the auxiliary branch supervised training band is used to further improve the performance of the network model. Given the prevalence of small-sized targets during the detection of multi-rotor UAVs, a small target replication and augmentation method is adopted. The target is first segmented, and then scaled at different ratios and placed in the background image to increase the number of small targets. Without relying on complex training techniques, our model has significantly enhanced the performance on the constructed UAV dataset compared to other detection models.