Estimation Of Forest Stock Volume Based On Feature Selection And Stacking Integrated Learning Algorithm

Forest stock volume is an important index to reflect the quantity of forest resources,and also an important forest stand survey factor.The purpose of this paper is to explore different feature selection methods and estimation models in order to provide new methods and ideas for the estimation of stock quantity in forestry science.In this paper,Boruta feature selection method,extreme gradient lifting algorithm and Stacking ensemble learning algorithm were first applied to the estimation of forest stock.Taking part of longquan city,zhejiang province as the research area,forest stock volume was estimated based on gaofen ii data,digital elevation model data and forest resources survey data.The main conclusions are as follows:(1)From high score 2 after pretreatment,respectively,for the four wave band information in remote sensing image: cross section,green wave band,and red band and near-infrared bands,the vegetation index is calculated by the band information,it is concluded that normalized differential vegetation index,ratio vegetation index,enhanced vegetation index,difference vegetation index and soil-adjusted vegetation index adjustment five vegetation index;After preprocessing the data of the digital elevation model,three topographic factors,such as elevation,slope and aspect,were obtained.The original feature set was composed of four field investigation factors including soil thickness,humus thickness,tree age and canopy density.(2)From the perspective of feature data sets,the accuracy of feature sets screened based on three feature selection methods(correlation analysis,stepwise regression analysis and Boruta feature selection)as the independent variable factor subset of the model is basically higher than that of all factors involved in the estimation.Boruta feature selection method is better than the other two methods.By comparing the estimation results of three machine learning methods,random forest method is the best,followed by extreme gradient enhancement method.(3)In this paper,the Stacking integrated learning model with the primary learner as extreme gradient boosting method,random forest method and gradient boosting method and the secondary learner as extremely randomized trees method is constructed.The results showed that the estimation results of the Stacking model constructed by Stacking method were above 82%,indicating that the Stacking integrated learning method had stronger generalization ability.At the same time,the estimation accuracy of the forest stock model constructed by Stacking ensemble learning method is better than that of various single algorithms,indicating that Stacking ensemble learning algorithm has important reference value for improving the estimation effect of forest resource stock.