YOLO-IRLight：基于YOLOv8的輕量級(jí)無(wú)人機(jī)紅外小目標(biāo)檢測(cè)算法
打開(kāi)文本圖片集
中圖分類(lèi)號(hào)：TP3191.4；TP301.6 文獻(xiàn)標(biāo)識(shí)碼：A
文章編號(hào)：2096-4706（2025）08-0046-09
Abstract：To address theisues of low detection accuracy and highcomputational load in infrared smalltarget detection from UAVaerial perspectives，a lightweight infrared smalltargetdetectionmodel，YOLO-IRLight，is proposed basedon YOLOv8s.Thismodelintroduces theEMA（EffcientMultiscaleAttention）Atention Mechanism toenhancefeatureextraction capabilities.APConv-C2f module isadded tothe neck of the network toreducecomputationalloadand fuse scale sequence features，andaP2detectionlayerisincorporatedtootimizethenetworkstructure，therebyimprovingsmaltargetdetection performance.Aovellgtweightdetectionead，Goup-Detect，isesigned，andtheNWD（NoaledussnWten Distance）lossfunctionisincorpoatedintothelossfunctionof themodelinalinearcombination maer.Experimentalresults on the open dataset show that compared to the original YOLOv8s， the proposed model improves detection accuracy （ m A P@ 0 . 5 ） （ by 1 . 7 % ，reduces the number of parameters by 4 5 . 9 % ，decreases computational complexity （GFLOPs） by 3 3 . 5 % ，and increases F1 score by 0 . 9 % .The improved algorithm significantly outperforms traditional algorithms，with notable improvements in detection accuracy compared to current mainstream algorithms.
Keywords： Small Target Detection; infrared target; lightweight; YOLOv8; network optimization
0 引言
目標(biāo)檢測(cè)是計(jì)算機(jī)視覺(jué)的重要任務(wù)，在各種領(lǐng)域如搜索救援、智能監(jiān)控中都有廣泛的使用。（剩余13247字）