改進(jìn)YOLOv8的無(wú)人機(jī)搜救小目標(biāo)檢測(cè)方法

打開(kāi)文本圖片集

DOI:10.16652/j.issn.1004-373x.2025.17.025 引用格式：,,，等.改進(jìn)YOLOv8的無(wú)人機(jī)搜救小目標(biāo)檢測(cè)方法[J].現(xiàn)代電子技術(shù),2025,48(17):167-175.

關(guān)鍵詞：無(wú)人機(jī)搜救；小目標(biāo)檢測(cè)；YOLOv8；特征融合；檢測(cè)頭；注意力機(jī)制中圖分類(lèi)號(hào)：TN911.73-34；TP391.4 文獻(xiàn)標(biāo)識(shí)碼：A 文章編號(hào)：1004-373X（2025）17-0167-09

nproved YOLOv8basedsmallobject detectionmethol forUAVsearchandrescue

JINSiyu,LIJiacheng,HUANGLan,CHENZhongju,ZHANWei (Collge of Computer Science,Yangtze University,Jingzhou 434023, China)

Abstract:Asmallobject detection model YOLOv8-CFTbasedonimproved YOLOv8 isproposed.It aims to improve the smallobjetdetectionperformance when UAVs performsearchandrescue misions incomplexoutdor environments.This model strivestoenhancefeatureextractioncapabilitiesbyintroducingtheC2F-SWCmodule.Meanwhile,itcombinesFADPN(feature aggregatinanddifusion pyramidnetwork)torealizeeficientfeaturefusion.Further,the TDHmoduleisusedtooptimize the detection head tobalance detectionaccuracyanddetectioneffciency.TheYOLOv8-CFTmodel wastestedonthe UAVsearch andrescuedataset UAVSRD.The YOLOv8-CFTmodel issignificantly beterthanthe benchmark YOLOv8 modelin termsof precision rate,recall rate and mAP,among which mAP @0.5 and mAP@0.5: 0.95 increase by 3.8% and 8.9%,respectively. ExperimentalresultsshowthattheYOLOv8-CFTmodelhasbetersmallobjectdetectioncapabilities inUAVsearchandrescue missions.

Keyword:UAVsearchandrescue;small object detection;YOLOv8;feature fusion;detection head;atention mechan

0 引言

無(wú)人機(jī)（UAV憑借其操作簡(jiǎn)便、靈活機(jī)動(dòng)的優(yōu)勢(shì)，逐漸成為復(fù)雜環(huán)境中執(zhí)行任務(wù)的理想工具，廣泛應(yīng)用于農(nóng)業(yè)監(jiān)控、災(zāi)害救援等多個(gè)領(lǐng)域-2]。（剩余15403字）