基于IRLT-YOLO的红外图像无人机目标实时检测研究OA

Unmanned aerial vehicles(UAVs)have been extensively employed across various fields;however,the increasing occurrence of unauthorized UAV"black flight"incidents poses a significant threat to public safety.In anti-UAV systems,infrared imaging sensors that are operational day and night are becoming increasingly prevalent in UAV detection and surveillance.This study addresses infrared image detection for UAVs and proposes a real-time infrared lightweight(IRLT)-YOLO target detection algorithm.In designing lightweight networks to reduce the depth of the backbone network,lightweight operations with a shared convolution in the header are employed,thereby minimizing redundant features.Real-time detection is achieved while preserving detection performance by introducing a tiny target detector-based on the normalized Wasserstein distance(NWD)-embedded in the loss function to replace the original Intersection over Union(IoU)metric.Experimental results indicate that the IRLT-YOLO model achieves precision,recall,mean average precision(mAP)@0.5,floating-point operations per second(FLOPs),and frames per second(FPS)of 95.4％,85.9％,89.5％,4.9G,and 167.0,respectively,representing a dual enhancement in computational accuracy and speed compared to the baseline model.Simulation experiments show that the IRLT-YOLO model enhances the detection and recognition capabilities of UAV targets in infrared scenarios,offering fast and effective real-time detection when deployed on edge devices,thereby meeting the demands of real-time detection applications in anti-UAV systems.