基于机器学习的静止卫星全天时云分类研究.pdf

Abstract

Therearemanykindsofclouds,whicharecomplexandchangeable.Differentcloudtypes

reflectdifferentatmosphericconditionsandevolutiontrends,indicatingfutureweatherchanges.

Forexample,deepconvectivecloudsareusuallyrelatedtosevereweather.Accurate

identificationofcloudtypesiscrucialtoimprovetheaccuracyofweatherforecastingandthe

effectivenessofclimatemonitoring.Geostationarymeteorologicalsatellitehastheadvantages

oftimecontinuity,highspatialresolutionandwidecoverage,whichcanprovidemoreabundant

spectralinformation.Therefore,automaticidentificationofsatelliteimageshasbecomeoneof

thehotspotsinthefieldofremotesensing.However,earlycloudclassificationmethodsmainly

relyonthefeaturedataofspectralchannelsforclassification.Therearestillsomeproblemsin

thesemethods.Forexample,thespectralcharacteristicsofdifferenttypesofcloudsaresimilar,

andthespectralcharacteristicsofthesametypeofcloudsmaybedifferent,whichleadsto

unsatisfactoryclassificationresults.Inaddition,earlystudiesmainlyfocusedonthe

classificationofdaytimeclouds.Duetothelackofvisiblelightbanddataatnight,itisdifficult

toobtainnighttimecloudcharacteristics,andcontinuousall-daycloudclassificationresults

cannotbeobtained.Therefore,theclassificationofnighttimecloudsisstillchallenging.

Geostationarymeteorologicalsatelliteprovidesalargenumberofcloudimagedata.

Miningandanalyzingthesedataandestablishinganeffectiveclassificationmodelarethekey

torealizeall-daycloudclassification.Basedonmachinelearningtheory,combinedwith

geostationarymeteorologicalsatelliteandCloudSatradarsatellitedata,thispaperstudiesthe