简介:Thedevelopmentofhigh-resolutionremotesensingimagingtechnologyprovidesanewwaytothelarge-scaleestimationofforestcanopydensity.Thetraditionalinversionmethodsforcanopydensityonlyusespectralortopographicalfeaturesofremotesensingimages.However,duetotheexistenceofthedifferentthingwithsamespectrumandthesamethingwithdifferentspectrumphenomena,itisdifficulttoimprovetheestimationaccuracyofcanopydensity.Basedonspectrumandothertraditionalfeatures,thispapercombinestexturefeaturesofremotesensingimagestoestimatecanopydensity.Firstly,thegraylevelco-occurrencematrix(GLCM)texturefeaturesarecomputedusingobjectbasedmethod.Then,theprincipalcomponentanalysis(PCA)methodisappliedincorrelationanalysisanddimensionreductionoftexturefeatures.Finally,spectrumandtopographicalfeaturestogetherwithtexturefeaturesareintroducedintostepwiseregressionmodeltoestimatecanopydensity.Theexperimentalresultsshowedthatcomparedwiththetraditionalmethodonlybasedonspectrumortopographicalfeatures,themethodcombinedwithtexturefeaturesgreatlyimprovedtheestimationaccuracy.Thecoefficientofdetermination(adjustedR~2)increasedfrom0.737to0.805.Theestimationaccuracyincreasedfrom81.03%to84.32%.