UavdNet:轻量化无人机图像目标检测网络

s drone technology rapidly evolves, its widespread applications in environmental monitoring, traffic management, and hazardous area reconnaissance have become increasingly significant. This study introduces a new lightweight Unmanned Aerial Vehicle image target Detection Network(UavdNet) suitable for edge devices like drones, enabling real-time embedded target detection tasks. UavdNet, built upon and optimized from the MobileNet network, incorporates an improved channel-space fusion attention mechanism, CASA, effectively addressing the complex backgrounds in drone images. Additionally, a new feature fusion module is designed to capture comprehensive feature information, enhancing the detection capability for small objects. Performance evaluations on the reorganized VisDrone datasets demonstrated that UavdNet maintains low computational complexity while achieving high detection accuracy, reaching peak accuracies of 67.1\% on the reorganized VisDrone datasets, respectively. This marks an improvement of 0.4\% over networks with similar parameter volumes.