Foggy　weather　is　often　encountered　on　the　battlefield,　and　environmental　visibility　will　be　reduced　when　conducting　operations　on　foggy　days,　which　often　affects　UAV　reconnaissance　of　armored　vehicles　on　the　ground.　To　improve　the　detection　accuracy　of　armored　vehicles　on　foggy　days,　this　paper　first　applies　the　AOD-Net　network　to　defog　the　foggy　images,　then　improves　the　YOLOv5　network　model　by　replacing　the　original　head　structure　with　the　DynamicHead　structure　and　uses　the　RepGFPN　pyramid　network　in　the　network　to　detect　the　target　in　the　defogged　images.　The　experimental　results　show　that　the　accuracy　and　recall　of　the　improved　network　model　are　increased　by　1%　and　3%　respectively,　and　the　mAP　is　increased　by　2.7%,　which　effectively　improves　the　accuracy　of　recognizing　armored　vehicles　in　foggy　images.　By　the　comparisons　between　the　original　images　and　the　defogged　ones　by　the　improved　model,　the　experimental　results　show　that　there　are　a　significant　increases　in　the　detection　accuracy　of　the　tank　targets　after　the　images　are　defogged.　©　2024　IEEE.