Study On Inversion Model Of Coniferous Forest Stand Volume Based On Texture Features Of Remote Sensing Images

In this paper Miyun County coniferous forest was for the study. The author discussed coniferous forest stock volume inversion model based on texture features by the same window or all the windows, and based on texture features by the same window and topographical factors, or texture features by all the windows and terrain factors. The best fitting model, the optimal combination of features and optimal texture window were filtered out. The author used TM image texture features to build inversion model of coniferous forest stand volume, which to provide a new research method and idea for prediction of forest stand volume.The author obtained the correlation between48texture feature extracted in the same window of TM remote sensing image texture and field stand volume by multiple regression method repeatedly, and built the model based on240texture characteristic factor generated from all remote sensing image texture window. Then, topographical factors were brought to model built, and as a result,6similar inversion models of forest stand volume combing texture features of remote sensing images with topographical factors. Reserved21fields described fitting accuracy of built models, by paired T-test or calculating the estimated accuracy of models. Radj-, RMSE, p-level paired T-test results and the model predictive accuracy were used as model predicting accuracy evaluation parameters.Radj2of each coniferous forest stand volume inversion model acquired based on different texture windows, such as3 × 3,5×5,7×7,9×9,11×11and texture feature generated from all the texture windows were0.411,0.324,0.351,0.369,0.331,0.860, respectively. Combined with topographic factors and texture Paired T-test with reserved21fields predicted that there is no distinct difference with testing results and predicted result of all built models. Moreover, forecast accuracy of inversion models were65.13%,59.52%,59.24%,65.46%,61.74%,81.36%, correspondingly.While combing topographic factors with texture parameters, The value of Radj2were0.748,0.342,0.442,0.342,0.402,0.890, and models’forecast accuracy were70.50%,54.78%,61.73%,54.78%,59.60%,82.13%, correspondingly.Studies have shown that the effectiveness and accuracy of fitting models based on texture features generated from all five groups were better than models based on those based on single texture window. And texture features of TM images can be used to coniferous forest stand volume estimation, with high degree of accuracy, also. In addition, combined topographic factors with TM remote sensing image texture features to build stand volume inversion model has greater accuracy than the models based only TM remote sensing image texture features. Therefore, models with topographic factors be considered will be with higher fitting accuracy. The best fitting model with R2adj=0.890, RMSE=9.885m3/hm2, Forecast accuracy P=82.13%is based on topographic factors and texture feature of all windows. There are some innovation points of the article:a) Applying TM Texture to forest stand volume inversion model, b) Combining texture features with topographical factors could improve the stock volume estimation accuracy further, c) Exploring affection of the texture window size with the accuracy of estimating stand volume.