基于频域增强与多尺度特征融合的无人机检测算法研究OA

In view of micro-small UAVs being met with a challenge in detecting against the complex background,this paper proposes an enhanced object detection framework based on the improved YOLOv12n(YOLO-Tiny-UAV).The algorithm is to introduce a wavelet transform-based frequency-domain feature decoupling module to strengthen feature extraction,construct a cascade detection head facing tiny UAVs to enhance feature representa-tion,and optimize bounding box regression accuracy in integration with a normalized Wasserstein distance(NWD)loss function.Meanwhile,a refined anti-UAV detection dataset(Anti-TinyUAV)is created,including multi-modal(infrared/visible light)data,multi-object configurations(single-target/swarm),and multi-aircraft labels(UAVs of 7 kinds),supporting intelligent decision-making for hierarchical countermeasure strategies.The experiments show that the model keeps an inference speed at 2.6 ms to achieve an mAP50 detection accuracy of 82.3％,an improve-ment of 5.65％over the baseline model,and a 4.58％increase at a recall rate for micro-small UAVs,providing an efficient solution for UAV monitoring in complex environments.