Merged Airborne LiDAR And Hyperspectral Data For Tree Species Classification In Puer's Mountainous Area

Forest ecosystem is the largest terrestrial ecosystem,related to people's lives.Accuratelyidentifing vegetation is the basis of studying forest resources by remote sensing technology.At present,the application of remote sensing technology in vegetation identification,what is mainly focused on the flat region where has a single tree species structure.There is a typical mountain landscape in Pu'er city,Yunnan province.Pu'er is located in the tropical and subtropical transition region.Influenced by subtropical monsoon climate,there are richness forest types and diverse vegetation.In the study,combined the advantages of multi-source data,based on the AISA Eagle II hyperspectral data and airborne LiDAR which were taken in April of 2014,derived height feature from point cloud data and spectral and texture features from hyperspectral remote sensing image.Then,the Principal Component Analysis(PCA)transformation was used to reduce the merged image.Finally,the Support Vector Machine(SVM)approach was used to identify and classify the main tree species of Pu'er city,thatprovide data support for ecological monitoring and sustainable management of forest vegetation.The main conclusions of the thesis areas follows:1.Based on TerraScan point cloud classification,digital surface model(DSM)and digital elevation model(DEM)are obtained,after subtraction to obtain a digital canopy height model(CHM)which can be a direct reflection of the vegetation height.2.The ACTOR4 modle was choosed to carry on the atmospheric correction for mountainous area image,and the Integrated Radiometric Correctionmodle is used to carry out topographic radiation correction,finally,getting the hyersperctral image which can more accurately reflect the features of ground object in the experimental area.3.The airborne LiDAR data and Eagle II AISA hyperspectral image can be merged which used the PCA transformation and based on the feature.The optimum index factor is 81.25,that shows the method is feasible as well as effective.4.SVM is introduced to the classification of fused image,how to select kernel function and parameter are the key in the classification,at last,by explored we determinated the RBF function as the kernel function,the penalty factor C=3,the parameter Gamma=0.008,meanwhile,analyzing the result of tree species classification.The result shows that the main tree species of study areaare Pinus kesiya,Betula alnoides,Castanopsis hystrix,Schima superba and so on,also,the total accuracy and kappa coefficient are 80.54%,and 0.78 in Wan Zhangshan,which are 8.85% and 11.43% higher compared with the classification accuracies without CHM;the total accuracy and kappa coefficient are 72.55%,and 0.69 in Cai Yanghe,which are 5.44% and 9.52% higher compared with the classification accuracies without CHM.5.The tree species classification image was posted process by Primary / Secondary Analysistechnique,in order to solve the problem about solitary point in classification.Finally,the matic map of classification was generated in study area.6.According to the classification results of the main tree species,the correspounding forest management measures were proposed with the two development goals of the experimentation area:(1)Pinus kesiya var.langbianensis plantationare are the main composition of Wan Zhangshan,through the creation of mixed conifer and broadleaved forest,to improve the forest structure,and realize the goal of sustainable development;(2)Cai Yanghe is natural forest region dominated by mixed broadleaf-conifer forest,vegetation is rich,the management mode of artificial promote natural regeneration is selected,to realize the forest multi-function goals.