基于YOL0v5模拟敌友智能判别算法

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）04-0044-05

Intelligent Enemy-friend Discrimination Algorithm Simulation Based on YOLOv5

YANG Hongli'， YU Hao²， JI Ruihang²， LI Wenzhuo²， YE Jinze²， LIANG Yuansheng² （1.School of Mathematics and Physics，Nanjing Institute of Technology，Nanjing ， China; 2.Engineering Training Center，NanjingInstitute ofTechnology，Nanjing，China）

Abstract：Toaddresstherecognitionchallengecaused bythe appearance similaritybetween enemyand friendly targets in militarysimulation environments，an intelligent discriminationalgorithmbasedonYOLOv5 is proposed.Amulti-dimensional tactical simulationdatasetisonstructedandathe-categoryaotationsystemofred，blueandblackisadoptedtodentify friendlyforces，enemy forcesand hostagesrespectively.Multi-sourcedataaugmentationandsamplebalancing techniquesare combined to improve dataquality.At thealgorithmic level，an atention mechanismand a multi-level feature fusion strategy areintroducedtoenhancethemodel'sabilitytodistinguishsimilartargets.Atthecomputationlevel，lightweightinference optimization isadopted to balance detection speedandaccuracy.Experimental results show thatthe model achieves anaverage precision mean （mAP@0.5） of 0.98 on the self-built dataset. It effectively realizes accurate identification and statistics of enemyandfriendlyunitsaswellashostagesandprovidesreliable technicalsupportforintelligentdeision-makingincomplex battlefield environments.

Keywords： YOLOv5 algorithm; Deep Learning; Object Detection; multi-feature fusion

0 引言

近年来，YOLO系列算法凭借其高效的实时检测能力，在军事目标检测领域中得到了广泛应用。（剩余5933字）