Study On Stand Type Identification Based On Airborne Hyperspectral And Lidar Data

Stand is the basic unit of forest investigation and mapping,and is the focus of forest resource management and monitoring.Hyperspectral remote sensing data contains dozens to hundreds of spectral channels,which can detect subtle spectral differences in ground objects.The differences in spectral information and texture information of different ground objects in remote sensing images can be used to distinguish different forest types,while lidar data can detect vegetation spatial structure and topography,making up for the defects of optical remote sensing.This paper takes Laoshan Shiye District of Maoershan Experimental Forest Farm of Northeast Forestry University as the research area.Firstly,the laser radar data(LiDAR)is preprocessed to remove redundant noise points and extract Digital Elevation Model(DEM).Then,the optimal segmentation scale is determined,multiscale segmentation processing is carried out on hyperspectral data,a classification system is determined,and the most representative sample is selected according to visual interpretation;34 features such as spectrum and texture are extracted from hyperspectral data and 112 features such as height and intensity are extracted from LiDAR data respectively.Using random forest feature selection method,the variables with higher importance in each feature are selected,and 11 feature variables are selected from hyperspectral and LiDAR data respectively,totaling 22 features.The selected feature variables are added to Random Forest(RF),Support Vecor Machine(SVM)and Gaussian Naive Bayes(GNB)classifiers respectively,and nine classification schemes are obtained by combining different feature variables and different classifiers for research.Finally,the classification results are compared and evaluated according to the overall accuracy(OA),user accuracy(UA),producer accuracy(PA)and Kappa coefficient.The specific conclusions are as follows:(1)Multi-source data can improve classification accuracy better than single data source.Compared with only using hyperspectral or LiDAR data,the classification accuracy of hyperspectral+LiDAR data source is improved by 2.45%and 8.22%,respectively.Among single data sources,the average classification accuracy of hyperspectral data sources is 80.47%,and that of LiDAR data sources is 74.70%,which is higher than that of LiDAR data sources.(2)The average classification accuracy of RF classifier is 82.92%,which is 81.19%higher than that of SVM classifier and 73.97%higher than that of GNB classifier.Among the three classifiers,the random forest classifier is superior to the support vector machine classifier,and the support vector machine classifier is superior to the Gaussian Naive Bayes classifier.(3)Chlorophyll index,as one of the biochemical parameters of vegetation,can effectively improve the classification accuracy of stand types,and the classification accuracy can be improved by about 3.32%after participating in the classification.(4)Among the five stand types,the classification effect of broad-leaved mixed forest is the best,while that of Korean pine forest is the worst.