With　the　rapid　development　of　flexible　DC　distribution　networks,　fault　detection　and　identification　have　also　attracted　people＇s　attention.　High-resistance　grounding　fault　poses　a　great　challenge　to　the　distribution　network.　The　fault　current　is　very　small　and　random,　which　makes　its　detection　and　identification　difficult.　The　traditional　overcurrent　protection　device　cannot　identify　and　act　on　the　fault　current.　Therefore,　this　paper　proposes　a　fault　detection　method　based　on　variational　mode　decomposition　(VMD)　combined　with　the　convolutional　neural　network　(CNN)　of　the　inception　module.　This　method　first　uses　VMD　to　decompose　the　positive　transient　voltage.　Second,　it　inputs　the　decomposed　signal　into　CNN　for　training　to　obtain　the　optimal　parameters　of　the　model.　Finally,　the　model　performance　is　tested　based　on　the　PSCAD/EMTDC　simulation　platform.　Experiments　show　that　the　detection　method　is　accurate　and　effective.　It　can　realize　the　accurate　identification　of　seven　different　fault　types.