YOLO-IRLight：基于YOLOv8的轻量级无人机红外小目标检测算法

打开文本图片集

中图分类号：TP3191.4；TP301.6 文献标识码：A

文章编号：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

目标检测是计算机视觉的重要任务，在各种领域如搜索救援、智能监控中都有广泛的使用。（剩余13247字）