Pinus yunnanensis volume estimation model based on UAV multispectral image
[Objective]Unmanned aerial vehicle(UAV)multispectral remote sensing images,with richer spectral information than visible light images,have great potential in forest volume estimation.Taking UAV-bome multispectral remote sensing images as the main data source,this study aims to explore the remote sensing estimation model of forest volume,so as to overcome the drawbacks of traditional ground survey,such as heavy workload,long time consumption and high cost.[Method]Taking the typical natural pure Pinus yunnanensis forest in Luomian Township,Fumin County,Kunming City as the research object,the single-band reflectance,vegetation index and texture feature were extracted according to the UAV multispectral image,and the standard ground mean of each characteristic variable was calculated.The characteristic variables significantly correlated with the forest volume were screened,and the forest volume estimation model was established using multiple linear regression,random forest and support vector machine.The model accuracy was evaluated by coefficient of determination(R2),root mean square error(ERMS),mean absolute error(EMA)and mean relative error(EMR).[Result](1)Among the three models,the random forest had the highest accuracy(R2=0.89,EMA=4.69 m3·hm-2,ERMS=5.45 m3·hm-2,EMR=14.5％),followed by the support vector machine(R2=0.74,EMA=5.27 m3·hm-2,ERMS=8.31 m3·hm-2,EMR=13.1％).The multiple linear regression model had the minimum accuracy(R2=0.35,EMA=10.12 m3·hm-2,ERMS=12.85 m3·hm-2,EMR=28.1％).The estimation accuracy of the three models in the test set decreased.The random forest had the best performance,followed by the support vector machine,and the multivariate linearity was the worst.(2)The three models had certain underestimation and overestimation in the estimation of P.yunnanensis forest volume.(3)Texture feature was still an important factor that could not be ignored in estimating the forest volume of P.yunnanensis based on UAV multispectral images.[Conclusion]Based on the multi-spectral images of UAV,the single-band reflectance,vegetation index,and texture factor mean values of the standard ground were extracted without individual tree segmentation,and the variables suitable for volume estimation were screened to construct an estimation model.Through the precision evaluation of the three models,the random forest is the best model for estimating P.yunnanensis volume.[Ch,2 fig.5 tab.27 ref.]
万方数据
维普
