Research On Beijing Forest Volume Estimation Model Based On Remote Sensing Texture Features

With the development of 3S technology,the method of combining remote sensing image data with ground survey data to establish a model for estimating volume has solved many deficiencies of traditional stock volume measurement methods and has been widely used.This study uses Beijing as the study area,combining GF-1 remote sensing image data,DEM data,and forest inventory data to explore the impact of remote sensing texture characteristics on the estimation of forest volume.At the same time,a multiple linear regression model,a BP neural network model,and a support vector regression model based on the three types of forest volume were established.By comparing the accuracy of the models established by three different methods,a more reliable method of estimating the accumulation is sought.The main research contents and conclusions are as follows:(1)The influence of the texture characteristics of remote sensing images on the estimation of volume is explored.Using the GF-1 satellite data as the data source,the spectral characteristics,topographical characteristics,and texture characteristics were extracted.Taking the broad-leaved forest in Beijing as an example,a multiple linear regression method was used to construct the volume estimation models without texture features and texture features,respectively,and the two models were tested for accuracy.The accuracy of the models with texture features in both models is higher than that of models without texture features.The determination coefficient R~2 of the accumulation quantity estimation model without texture features is 0.52,and the root mean square error RMSE is 10.51m~3/ha;the determination coefficient R~2 of the accumulation quantity estimation model with texture features added to a 3×3 window is 0.55,and the root mean square error RMSE is 10.34 m~3/ha.(2)The influence of different texture windows of high-score images on the estimation of forest volume is explored.Taking Beijing broad-leaved forest as an example,texture features based on five different windows(3×3,5×5,7×7,9×9,15×15)were extracted and combined with spectral features and terrain features,respectively.Based on the multiple linear regression method,five accumulation volume estimation models based on different window texture features are constructed,and the five models are tested for accuracy.The results show that the accumulative volume estimation model based on 7×7window texture features among the five models has the highest accuracy,the determination coefficient R~2 is 0.56,and the root mean square error RMSE is 10.22m~3/ha.(3)The stock volume estimation models of three forest types in Beijing based on texture features were constructed.Combining spectral features extracted from high-scoring remote sensing images,7×7 window texture features,and topographic features,using traditional multiple linear regression methods and two machine learning-based BP neural network algorithms and support vector regression algorithms to build Beijing,respectively Accumulation estimation models of coniferous forest,broad-leaved forest and mixed forest in the city,and the accuracy of the models established under the three methods is tested.The results show that among the three modeling methods,the accuracy of the SVR model based on machine learning is better than the BP neural network model and the traditional multiple linear regression model.Among the three forest types,the mixed forest volume estimation model is superior to the broad-leaved forest and coniferous forest volume estimation models.The mixed forest SVR model has a determination coefficient R~2 of 0.75 and a root mean square error RMSE of11.02 m~3/ha.