Global　warming　has　exacerbated　the　melting　of　glaciers　in　Greenland　and　Antarctica,　which　has　considerable　influence　on　sea　level　rise.　The　classification　of　targets　in　the　radar　images　acquired　by　airborne　radar　is　helpful　to　further　analyze　the　evolution　of　ice　sheet　subsurface　targets.　In　recent　years,　deep　learning　techniques　for　target　detection　and　classification　greatly　improved　the　performance　of　traditional　techniques　based　on　hand-crafted　feature　engineering.　Therefore,　this　paper　proposes　a　fully　convolutional　network　for　the　automatic　target　classification　of　ice　sheet　radar　image,　which　is　mainly　composed　of　feature　extraction　module　(FEM)　and　densely　dilated　convolution　module　(DDCM).　The　FEM　combines　standard　convolution　and　depthwise　separable　convolution　together,　which　can　reduce　the　parameters　and　maintain　the　classification　accuracy　as　much　as　possible.　The　DDCM　is　composed　of　four　dilated　convolutions　with　different　dilation　rates　by　dense　connection,　which　is　used　to　capture　larger　receptive　field　and　obtain　high-level　semantic　features.　The　proposed　network　is　validated　on　ice　sheet　radar　image　provided　by　the　Center　of　Remote　Sensing　of　Ice　Sheets　(CReSIS)　from　2009　to　2011.　Experimental　results　show　that　the　proposed　network　achieves　higher　accuracy　than　the　classic　backbone　networks,　and　it　is　about　8　times　faster　than　the　latest　ice　sheet　image　target　classification　network　to　infer　each　image.　©　2021　ACM.