Vector　vortex　beams　(VVBs)　are　a　promising　type　of　structured　light　that　combine　the　orbital　angular　momentum　(OAM)　and　the　polarization　states　of　light.　Due　to　their　intrinsic　high　dimensionality,　VVBs　show　great　advantages　in　applications　like　optical　communications,　information　encryption,　and　quantum　information　processing.　However,　the　high　dimensionality　presents　a　challenge　for　pattern　detection.　In　this　paper,　we　compare　different　machine　learning-based　methods　for　classifying　270　classes　of　VVB　using　basic　CNN,　MobileNet,　and　ResNet18　neural　networks.　We　visualize　the　VVB　modes　using　a　color-coding　method　with　Stokes　parameters,　and　the　neural　networks’　performance　is　tested　in　a　1　km　free　space　communication　link　with　four　atmospheric　turbulence　strengths.　The　results　demonstrate　that　neural　networks　can　recognize　large　datasets　of　laser　modes　with　good　accuracies,　even　under　turbulence　environments.　We　also　propose　an　image　encryption　scheme　using　the　VVB　dataset　to　encode　an　RGB　figure　which　is　transmitted　through　the　turbulence　channel　and　successfully　recovered　by　the　pre-trained　neural　networks.　Our　study　highlights　the　potential　of　artificial　intelligence　for　VVB　pattern　recognition　and　could　have　a　significant　impact　on　the　design　of　future　optical　communications　systems　and　information　encryption　protocols.　©　2023　Elsevier　GmbH