With　the　acceleration　of　intelligent　transportation　in　the　new　era,　license　plate　recognition　technology　has　become　more　and　more　extensive.　Although　the　existing　license　plate　recognition　technology　can　show　good　performance　for　specific　scenes　and　traditional　license　plate　recognition,　it　has　poor　recognition　robustness　and　applicability　in　complex　external　environments　such　as　arbitrary　shooting　angles,　different　lighting　conditions,　and　different　picture　quality.　To　solve　the　above　problems,　this　paper　proposes　an　improved　recognition　algorithm　suitable　for　complex　scenes　and　various　types　of　license　plates.　First,　the　SSD　model　is　used　for　multi-target　vehicle　detection;　the　convolutional　neural　network　(CNN)　is　further　used　for　license　plate　location　and　correction;　finally,　the　YOLOv2　model　with　modified　output　and　reduced　pooling　layer　is　designed　to　realize　the　license　plate　character　recognition.　The　algorithm　in　this　paper　is　carried　out　on　the　surveillance　image　data　of　a　road　in　a　block　of　Beijing,　which　contains　3016　images　with　a　total　of　5315　vehicles　to　be　detected.　The　experimental　results　show　that　the　accuracy　rate　of　the　license　plate　recognition　of　the　proposed　method　reaches　95.66%,　and　it　has　considerable　detection　performance　for　traditional　license　plates　and　new　energy　license　plates.　Compared　with　the　classic　algorithm　HyperLPR,　the　recognition　accuracy　of　the　proposed　algorithm　in　complex　scenarios　is　improved　by　13　%.　©　2020　ACM.