Tongue　image　texture　analysis　is　the　main　content　of　inspection　diagnostics　in　traditional　Chinese　medicine　(TCM).　The　tough　and　tender　classification　for　tongue　image　is　mainly　represented　by　image　texture.　The　traditional　extraction　method　of　texture　feature　is　not　robust　and　is　not　suitable　for　variable　illumination　and　different　image　acquisition　equipment.　So　we　select　＇Xception＇　as　the　main　network　model　to　complete　feature　extraction　firstly.　Then,　two　full　connection　layers　and　a　softmax　layer　are　combined,　two　dropout　layers　are　added　in　their　middle　to　form　a　classification　network.　The　experimental　results　show　that　the　better　accuracy　rate　can　be　obtained　by　our　method.　©　2020　ACM.