In　the　process　of　construction　and　operation　of　mountain　roads,　slope　disasters　such　as　landslide　and　collapse　are　often　encountered,　which　seriously　affect　the　transportation　infrastructure　and　safe　operation　in　China.　Cracks　are　the　early　symptoms　of　most　slope　diseases.　By　monitoring　the　change　trend　of　cracks,　the　displacement　trajectory　of　the　slope　body　can　be　reflected　in　time,　which　is　of　great　significance　for　landslide　monitoring　and　early　warning,　so　the　safety　detection　is　concentrated　in　this　stage.　In　recent　years,　great　progress　has　been　made　in　deep　learning-based　computer　vision　methods,　which　have　the　advantages　of　simple　observation　method,　low　cost,　wide　detection　area　and　sustainable　monitoring.　In　view　of　this,　a　pixel　level　segmentation　method　of　slope　cracks　based　on　deep　convolutional　neural　network　is　proposed　in　this　paper.　According　to　the　shape　characteristics　of　slope　cracks,　a　deep　convolutional　neural　network　was　designed.　The　network　was　trained　on　the　self-made　slope　image　data　set,　and　the　IOU　on　the　validation　set　reached　75.26%,　which　realized　the　precise　segmentation　and　recognition　of　cracks.　Experimental　results　show　that　the　model　has　a　good　ability　to　characterize　the　slope　cracks,　can　accurately　extract　the　slope　cracks,　and　provides　a　reliable　basis　for　the　formulation　of　slope　early　warning　and　disaster　relief　programs.　©　2021,　Springer　Nature　Switzerland　AG.