红外与可见光多模态融合检测网络YOL0-MF及其恶劣环境鲁棒性研究

打开文本图片集

中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2026）04-0086-08

Research on YOLO-MF Multi-modal Fusion Detection Network for Infrared and Visible Lightand Its Robustness in Harsh Environments

MATenglong，CHEN Yanlin，LIJiaqiang，YUHaisheng，HE Chao （College ofMachinery and Transportation， Southwest Forestry University， Kunming 650224， China

Abstract：This paper proposesamulti-modalfusionmodel，YOLO-MF，for infraredandvisiblelighttoadressthe issues of insufficientobject detectionaccuracycaused byilumination fluctuations and harsh weatherincomplex trafic scenarios. This modelis builtuponYOLOvllandconstructsa“dualinput-multiscale interaction-end-to-end detection”architecture. It integrates threefusion modules includingBiFocus，AIFI，andDEAas well astheEUCBdetection head toachieve noise suppression，dynamic interaction，andfeature enhancement.Experimental validationresultsonthreemajordatasetsshowthat the mAP（∅0.5 onthe LLVIP test set reaches 0.964 and the m reaches O.629，with a parameter count of only 7.5M Inlow-lightscenaros，theprecisionis0.944andtherecallis0.918，whichissignificantlysuperior tomastreammodelssuch astheYOLOseries （v5-v12）andRT-DETR.The modeladapts topurevisiblelight，pure infrared，and multi-modalcomplex scenarios.Thismodelefectivelysolves thepain pointsofpoorsingle-modalrobustnes，noise interference intraditionalfusion， andinsuffcientreal-time performance，providing ahigh-precision detection solution forfelds suchasautonomous driving.

Keywords：YOLOv11; BiFocus;DEA; EUCB;AIFI; multimodal fusion

0 引言

目标检测作为计算机视觉的核心任务，在自动驾驶等领域发挥着不可替代的作用。（剩余11158字）