Accurate　and　fast　recognition　of　power　quality　disturbances　(PQDs)　is　very　significant　for　power　pollution　control.　A　novel　method　based　on　visualization　trajectory　circle　(TC)　and　machine　vision　is　proposed　to　ameliorate　the　recognition　accuracy　of　complex　PQDs.　To　obtain　the　anti-interference　stationary　analytic　signal　sequence,　an　improved　Hilbert　transform　(IHT)　is　performed　on　single　and　complex　PQD　signals.　The　instantaneous　amplitude　and　phase　are　taken　as　polar　radius　and　angle　to　obtain　the　TC　image　in　polar　coordinates.　The　images　are　input　in　ResNet50　for　training　to　achieve　the　optimal　network　model,　to　realize　the　type　recognition.　Finally,　the　proposed　method　is　tested　by　the　synthetic　database,　which　is　built　from　mathematical　models　and　compared　with　other　advanced　methods.　In　addition,　time　interval　detection　can　be　realized　by　the　Hilbert　spectrum　based　on　IHT.　Simulation　results　demonstrate　that　the　method　has　strong　robustness　and　high　accuracy.　Furthermore,　a　13-node　microgrid　test　system　with　distributed　generations　is　built　on　the　RT-Lab　platform,　to　generate　PQDs　for　further　validating　the　method.　The　single　or　complex　PQDs　caused　by　test　events　are　successfully　recognized.