改进YOLOv8的雾天目标检测算法：BRES-YOLO

打开文本图片集

DOI：10.16652/j.issn.1004-373x.2025.17.013 引用格式：，等.改进YOLOv8的雾天目标检测算法：BRES-YOLO[J].现代电子技术，2025，48（17）：85-92.

关键词：目标检测；雾天驾驶；YOLOv8；BiFormer；多尺度注意力机制；C2f-RepGhost中图分类号：TN911.73-34；TP391.4 文献标识码：A 文章编号：1004-373X（2025）17-0085-08

Improved YOLOv8 based foggy object detection algorithm： BRES-YOLO

MAOHanwei，LIShixin，ZHOULiming，WANGPeng，ZHU Zhiren （SchoolofElectronicEngineering，TianjinUniversityof TechnologyandEducation，Tianjin3oo22，China）

Abstract：Visibilityisreduced infoggyscenarios，andtheobjectinformationcapturedbyvehiclesinfoggyweatherisvague andincomplete，soitispronetofalsedetectionandmissddetection.Inviewof this，animprovedYOLOv8basedfoggyobject detectionalgorithmBRES-YOLOisproposed.FirstlyBiFormerisintroducedtoreplacethebackbonefeatureextractionetwork tocapturediferentsemanticfeatures，sostoenhancetheetectionaccuracyofthemodel.SecondlyanEMA（eficientmultiscaleattention）moduleisaddedtotheneck network toenhance the model'satentiontotheobject.TheSPD-Conv（space-todepthconvolution）isintroducedtoreplace theConvintheoriginalmodel toimprovethedetectionperformancewhendealing withimages withlowresolutionandsmallobjects.TheMPDIoU（minimumpointdistancebasedIoU）isusedtoreplace the originallossfunctiontoimprovethelocalizationauracyandclasificationperformanceofthedtectionframe.FinalytheCfRepGhoststructureisintroducedtoreplacetheoriginalC2f moduletolightenthemodelandreduce thecomputational complexityofthemodel.Theexperimentalresultsshowthatthe meanaverage precisionmAP@0.5of BRES-YOLO modelon RTTSdatasetis increasedby5.8%，anditsmAP@0.5：0.95isincreased by5.4%incomparisonwiththoseof YOLOv8n.In summary，the BRES-YOLO model can accomplish the task of object detection in foggy scenes more accurately.

Keywords：objectdetection;driving infoggyweather;YOLOv8; BiFormer;multi-scale atention mechanism; C2f-RepGhost

0 引言

随着科技的飞速发展，目标检测逐渐成为自动驾驶、无人机控制领域的关键技术。（剩余12808字）