The　wind　turbine　blades　are　the　key　part　of　converting　wind　energy　into　electrical　energy.　Currently　the　fault　diagnosis　of　blades　is　mainly　dependent　on　manual　visual　inspections.　In　this　paper,　an　image　based　fault　diagnosis　method　for　wind　turbine　blades　is　proposed.　The　blade　damage　recognition　is　realized　by　two-stage　learning.　The　first　learning　stage　is　deep　feature　extractor　learning　stage.　A　deep　convolutional　neural　network　(ConvNet)　is　built　and　trained　on　the　ILSVRC　dataset,　owing　to　the　lack　of　the　blade　damage　images.　The　trained　ConvNet　without　the　last　two　layer　is　extracted　as　the　feature　extractor　of　the　blade　images.　The　second　learning　stage　is　pattern　learning　stage.　Deep　features　of　the　training　blade　images　are　extracted　and　used　to　train　a　classifier　to　identify　the　damage　type　of　the　blades.　The　damage　identification　of　the　wind　blades　can　be　realized　by　the　combination　of　the　deep　feature　extractor　and　the　classifier.　The　wind　turbine　blade　images　in　the　experiment　were　taken　at　a　real　wind　power　station.　Three　conventional　feature　extraction　methods,　which　are　Histogram　of　Oriented　Gradient,　Scale　Invariant　Feature　Transform　and　Texture,　are　adopted　to　compared　with　the　proposed　method.　The　effectiveness　of　the　proposed　approach　is　validated　through　experiments,　and　the　results　show　that　for　the　damage　recognition　of　wind　turbine　blades.