Cervical　cancer　is　a　major　threat　to　women’s　health　and　there　is　a　huge　population　suffering　from　it　in　the　world.　Colposcopy　screening　is　one　of　the　important　methods　for　early　diagnosis　of　cervical　cancer.　In　this　paper,　we　propose　a　method　based　on　deep　learning　for　colposcopy　images　recognition,　which　could　be　used　for　early　screening　of　cervical　cancer.　The　method　is　mainly　composed　of　two　parts,　the　segmentation　of　the　diseased　tissue　in　the　colposcopy　image　and　the　classification　of　the　image.　In　our　method,　the　U-Net　is　used　to　extract　the　ROI　of　images　and　a　deep　convolutional　neural　network　is　designed　to　extract　features　for　classification　of　the　ROI.　In　addition,　we　introduce　the　spatial　attention　mechanism　to　make　the　neural　network　pay　more　attention　to　the　diseased　tissue　in　images.　Experiments　demonstrate　that　the　proposed　method　has　a　good　performance　on　the　colposcopy　images,　and　even　achieve　nearly　test　accuracy　of　68.03%,　which　is　better　than　others　by　∼6%.　©　Springer　Nature　Switzerland　AG　2019.