Study On Information Extraction Methods For Cluster Bamboo Forests In Southwest Yunnan Based On Sentinel-2

There are abundant bamboo forests and bamboo species in Southwest Yunnan,which play an important role in local ecological protection and economic development.The bamboo in this area is widely distributed in mountainous areas,with scattered distribution,difficult investigation and lack of basic data,which limits the scientific planning and rational development and utilization of bamboo forests.Remote sensing is an important means to acquire resource information objectively,efficiently and cost-effectively.At present,there are few systematic studies on cluster bamboo using remote sensing technology,and the key technologies such as classification method,information feature and data selection are still unclear.Therefore,cluster bamboo in it area as the research object,Sentinel-2,Google Earth image as data,research on cluster bamboo information extraction,and combining the DEM,field survey data,illuminates the quantity and spatial distribution characteristics of bamboo.This study provides basic data for data information management,updating and development decision of bamboo industry,and provides technical support for remote sensing technology to carry out survey of cluster bamboo forests.The main conclusions of this study are as follows:1.Comparative study on classification methods of cluster bamboo forestBy comparing the random forest,back propagation neural network,support vector machine(SVM)as a result,the extraction of three kinds of size classification precision is respectively:the overall accuracy of 90%,78%,82%,Kappa coefficient of 0.87,0.72,0.78,cluster bamboo producers accuracy is 81%,65%,81%,user accuracy of 81%,61%,71%,random forest in classification precision and efficiency is better than the other methods.2.Information extraction of cluster bamboo forest with multi-feature optimizationUsing the Sentinel-2 image of December as the data source,four feature sets were constructed,including spectrum,vegetation index,red-edge vegetation index and texture.Five combinations and multi-feature optimization methods,and the random forest method was used to extract the information of bamboo forest.The results show that the red-edge vegetation index feature set can improve the accuracy,all four types of feature sets in the classification will result in feature redundancy,and the optimal accuracy cannot be obtained.The multi-feature optimization method reduced the number of features by 60.7%and obtained the best feature collection,which was composed of 6 spectral bands,3 red-edge vegetation indexes and 2 texture features.Compared with other schemes,multi-feature optimization improves the precision obviously,with the overall accuracy reaching 93.2%,Kappa coefficient up to 0.89,and the precision of cluster bamboo users up to 88.04%and producers up to 85.43%.Compared with the spectral features only,the overall accuracy increased by 3.2%,Kappa coefficient by 0.02,the accuracy of cluster bamboo users by 7.32%,and that of producers by5.76%.Multi-feature optimization can not only effectively reduce redundant features in classification,but also improve classification accuracy and efficiency.3.Study on optimal time phase of information extraction of cluster bamboosUsing multiple time phase Sentinel-2 images,four feature sets were constructed,including spectral bands,vegetation index,red-edge vegetation index and texture,which were optimized by different phases and multiple features.Random forest classification method was used to compare the extraction accuracy and difference in feature importance of different phases.The results showed that for the producer precision of bamboo forest information extraction,PA(March)>PA(February)>PA(December)>PA(November)>PA(April)>PA(January)>PA(May),UA(March)>UA(November)>UA(December)>UA(January)>UA(April)>UA(February)>UA(May),showed the optimal overall classification accuracy and Kappa coefficient in December,which were 93.2%and 0.89,respectively.In March,the accuracy of cluster bamboo users and producers was the best,89%and 86.95%,respectively.More images from December to March should be used in the follow-up study.In the result of multi-feature selection,red edge vegetation index,spectral feature,vegetation index and texture feature account for 41.97%,37.04%,16.05%and 4.91%,and red edge and spectrum play an important role.The feature set of red edge vegetation index combined with multi-feature optimization method should be used to extract remote sensing information of tuber bamboo.4.Mapping of cluster bamboo forests in Southwest Yunnan and distribution characteristicsThe mapping and spatial distribution characteristics of cluster bamboo forest were studied based on the information extraction results of December,DEM and Google Earth images.The first map of cluster bamboo forest in Cangyuan county with a resolution of 10m was drawn,covering an area of 138.07km~2.The main distribution types are four-side bamboo forest,pure bamboo forest and protective bamboo forest.four-side bamboo forest are mainly distributed around towns and villages in a ring or sporadic distribution.Protective bamboo forest is mainly distributed around cultivated land,shrub and grassland.Cluster bamboo is mainly distributed in the elevation of 900m～2000m,accounting for 93%of the total area.The slope is mainly distributed in gentle slopes and slopes,accounting for 90%of the total area.Using the result of multi-feature optimization of December Sentinel-2 image as the data source,the bamboo forest distribution map with a resolution of 10 m was drawn for Cangyuan County,and the total area of the clustered bamboo forest in Cangyuan County was 138.07km2.Combined with DEM data,GIS statistical analysis method was used to study the topographic distribution characteristics of bamboo forest.It was found that the fasciculus bamboo forest in this county was mainly distributed at an altitude of 900 m～2000 m,with a distribution area of128.4 km~2,accounting for 93%of the total area.Bamboo forests are mainly distributed on gentle slopes and slopes,covering an area of 127.3km~2,accounting for 90%of the total area.In combination with Google Earth,cluster bamboo forest spatial distribution characteristics of the study,bamboo forest type has sides bamboo forests,protective bamboo,and plantation,its sides bamboo distribution in surrounding towns and villages,assumes the circular or sporadic distribution,protective bamboo forest distributed in cultivated land,shrub and grass around,plantation small area distribution in the town of Mengsheng.This study proposed an efficient and low-cost remote sensing monitoring technology for the information of bamboo resources,and illustrated the number and spatial distribution characteristics of bamboo in Cangyuan County.The research results provided basic data support for the development of local bamboo resources and characteristic bamboo industry.