| The city is the most important place of human production activities,the most important areas of carbon emissions.The concentration of carbon dioxide in urban air is several times higher than that in urban areas,the proximity effect will make urban greening carbon fixation efficiency is relatively high.Therefore,it is very important to monitor and analyze the status of urban green space for the management and evaluation of urban green space.In this paper,we used the method of object-oriented multi-scale segmentation to extract the urban Green space information of the study area by using spectrum,shape and texture base on the WorldView-2 of high-resolution remote sensing image in West Lake District of Hangzhou.And then used random forest regression model was established to estimate the vegetation green biomass based on the biomass data of survey in the West Lake District.Finally analyzing the spatial variation characteristics.The main research contents are as follows:(1)According to the differences feature of the WorldView-2 high-resolution remote sensing image of the study area,divided into different multi-scale segmentation methods,constructed the multi-level structure,Comprehensive utilization of spectral,shape and texture features to extract urban green space information.In this paper,J-M(Jeffries-Matusita)distance method is used to determine the texture window scale and texture features.Finally,the CART(Classification And Regression Trees)decision tree classification algorithm is used to select the optimal feature and the node threshold region to classify the Land use type.(2)The survey data were divided into natural forest land and artificial green land.The biomass model was used to calculate the biomass.The correlation coefficient of remote sensing image was extracted by using the size of the sample.The correlation analysis was made with the biomass of the urban green space as the dependent variable.The random forest regression model was constructed in the sub-region,and the city of Hangzhou West Lake District was estimated by object-oriented extraction Green space vegetation biomass.Finally,according to the results of biomass estimation,the spatial distribution pattern of sub-area and sub-township streets and different hatching lines were analyzed.The main conclusions are as follows:(1)In this study,the texture scale and dimensionality reduction features were selected by using the separable exponential J-M distance method.The optimal texture window dimensions of the grassland,agricultural land,shrub and tree were 5 × 5,11 × 11,13 × 13,13 × 13.In addition,the texture features are reduced from 192 to 34 texture bands,which improves the accuracy and efficiency of information extraction.(2)The sub-regions are divided into different multi-scale segmentation methods,and the multi-level structure is constructed.The CART decision tree classification algorithm is used to quickly and effectively determine the characteristics of each hierarchical classification and node threshold by using the characteristic variables such as spectrum,shape and texture.The classification rules were used to extract the urban green space information from the study area.The results show that the average user accuracy of this object-oriented green space is 84.63%.Which is higher than 72.73% of the maximum likelihood method based on the pixel.The overall classification accuracy is improved from 76.53% to 88.56% and the Kappa coefficient from 0.7117 to 0.8623 from the maximum likelihood of the pixel-based method.Object-oriented CART decision tree taxonomy is not only superior to traditional pixel-based methods in many respects,but also more flexible and efficient use of spectral,shape and texture features than other researchers using object-oriented methods.Classification accuracy.(3)The trees in the study area were divided into three sub-regions: urban area,suburban area,forest area.And the three regions were plotted to estimate the biomass respectively.The shrubs were used to estimate the biomass according to the regression analysis of the plot data.The grassland was estimated according to the unit area.The Modeling fitting precision of urban area,suburban area,forest area and shrubs were 77.54%,74.38%,81.29% and 84.51%.The prediction accuracy was 71.32%,68.59%,76.08% and 80.80%.So the regression modeling accuracy can meet the experimental study.(4)The total biomass of urban green space vegetation in Hangzhou West Lake District was 151.8146 × 104 Mg,and the biomass density was mainly distributed in the range of 15-170 Mg / hm2,and the average biomass density was 49.36 Mg / hm2.The density of vegetation biomass in each area of the West Lake District is forest area> urban area> suburb area.The average density of different towns and villages in the order of the average density of trees from the largest to the order of the West Lake Street> LingYin Street> LiuXia street> Beishan Street> ZhuanTang Street> ShuangPu Town> GuDang Street> JiangCun Street> SanDun town > Xixi Street> Wen Xin Street> CuiYuan Street.Analysis of profile of biomass in West Lake District :The East-west direction is the highest level of biomass density;The North-South streets of West Lake District average biomass density level is relatively high,and the other location in the city area,agricultural land on the suburb area of the average biomass density is relatively low;The southwest-northeast direction ShuangPu town and West Lake Street average biomass density level is relatively high;The northwest-southeast,LiuXia street and West Lake street average biomass density is relatively high.Through the section line analysis and the analysis of space layout,the West Lake street,Zhuan Tang Street and LiuXia street average biomass density and biomass are high. |
PDF Full Text Request