An Enhanced YOLOX-Tiny Model for Unmanned Aerial Vehicle Signal Detection and Recognition

Addressing the challenge of UAV signal detection and recognition in complex scenarios characterized by multiple targets and co-frequency interference, this paper proposes an enhanced YOLOX-Tiny model. The model incorporates a Coordinate Attention (CA) mechanism, Depthwise Separable Convolution (DSC), and a Slim-Neck structure to reduce computational complexity while enhancing feature extraction capability. Furthermore, the Focal-EIoU loss function is introduced to mitigate the issue of imbalanced target samples, thereby improving detection accuracy and convergence speed. Experimental results demonstrate that, compared to the baseline model, the improved model achieves a significant increase in mean Average Precision (mAP) by 3.76%, reaching 87.50%, an enhancement in detection Frame Per Second (FPS) by 12 frames to 57 FPS, and an improvement in parameter estimation accuracy by 1.69%. With its high accuracy and lightweight nature, the proposed model offers an effective solution for real-time UAV signal detection and recognition in complex environments.