Study On Algorithm Of Forest Volume Estimation By High Resolution Remotesensing

The forest,with its abundant resources and powerful functions of oxygen production,providing the basic condition to survival for human beings.Its healthy growth plays an important role in the development of human society.The rapid,real-time and accurate control of forest stock volume has an important role for forest resource management.This paper takes GF-1 satellite image as data source,using the calibration parameters by Chinese Resources Satellite Application Center released on 2013.Doing radiometric calibration,atmospheric correction and ortho correction,image clipping processing on the 16m multi spectral image of the study area.Through the extraction and selection of the remote sensing image spectral information,texture information and terrain information,using the principal component analysis,partial least squares and random forest method to construct forest volume estimation model,then estimating the forest volume in study area.Through analysis the accuracy of the estimation of the large area and the small size of the study area,discussing the advantages and disadvantages of each method,the conclusion as followed:(1)Good image preprocessing results have important significance-for forest stock estimation later.(2)Join the texture as forest volume estimation factor has significant effect than the just use the spectral information and terrain information as the forest volume estimation factor.The accuracy of estimation result without texture information is 84.3%,the result with texture information has a better accuracy which accuracy is 87.3%.(3)In the global forest stock volume estimation results,random forest volume estimation accuracy is the best,up to 88.2%,partial least squares estimation results up to 87.7%and the principal component analysis of the estimation results for the 87.3%which is lowest.(4)In the small size volume estimation results,the result accuracy of random forests is the best which up to 75.33%,partial least square method precision is 73.24%,the principal component analysis method to estimate the results for the 73.18%lowest.the last but not least,a simple program of the core algorithm of the random forest regression is realized,which provides a reference for the application of random forest regression.