Classification Of Forest Type Using Hyperspectral Remote Sensing Data

Forests,grasslands,wetlands and other terrestrial ecosystems mainly take atmosphere and soil as media,providing a variety of living environment for humans and animals,and the main part of food and clothing.They play a decisive role in the survival and development of human beings.In the investigation of land ecological resources,trees,grasses,and water is the focus of the investigation,many ecological parameters are relying on the trees and grass seed and water to quantitatively estimate,therefore,accurately identify species,grass and water is the key to obtain the ecosystem information.The application of remote sensing technology to fine classification and target recognition of ecosystems can provide information services for natural resources management,environmental protection and monitoring,biodiversity and wildlife ecological status investigation.Spaceborne hyperspectral remote sensing image has high spectral resolution,and it is the main technical means to obtain the fine structure related parameters of forest ecosystems such as forests,grasslands and wetlands.This dissertation comprehensively uses the spectral information,texture information and terrain information of spaceborne hyperspectral remote sensing images,and explores the fine classification methods and models of forest ecosystems based on Hyperion satellite-borne hyperspectral data,and establishes corresponding fast and fine classification systems for forest ecosystems.Among them,the fine classification methods and models for forest ecosystems provide robustness for China’s GF-5 satellite-borne hyperspectral remote sensing data in natural resource management,environmental protection and monitoring,biodiversity,and wildlife ecological status surveys.Efficient image fine classification and target extraction technology.The main research results and conclusions of this paper are as follows:(1)In the feature-band-optimized dimension reduction method,the feature variables selected by the adaptive band index method outperform the random forest feature selection method in the C5.0 decision tree algorithm forest classification,and the overall classificationaccuracy of the two methods,reached 84.04% and 78.11% respectively.In the three dimensionality reduction strategies based on feature extraction,the accuracy of independent principal component analysis is high,and the accuracy of kernel principal component analysis is the lowest.Principal component analysis is between two methods.(2)Apply C5.0 decision tree data mining algorithm,adopt hierarchical classification Strategy,add relevant texture features and terrain factors on the basis of the spectral features of the integrated image,can effectively improve the precision of forest type classification,the method of forest type Fine classification achieves the dominant tree species(group)level,enabling semi-automation of the fine classification of hyperspectral data forest types.(3)Based on the combined model of adaptive weighted multi-classifiers,combining the support vector machine and random forest two kinds of machine learning algorithms at the pixel level,the precision of the forest type fine classification is effectively improved.