Feature-based Classification Researches In Hyperspectral Databases

Hyperspectral remote sensing plays an important role in monitoring global environment, guiding agricultural production and geological exploration field, spectral data mining is the premise and foundation of remote sensing applications. In the process of spectrum data acquisition, spectral data usually have a certain degree of error, as the affection of environment factors. It is necessary to do data processing. Meanwhile with the development of remote sensing technology, spectral data shows the characteristics of multi-band, large amount of data, so we should do features extraction, and use the characteristic parameters extracted classify. This paper discussed the data preprocessing and feature extraction based on the characteristics of hyperspectral imaging, also researched on methods of Decision tree classification and principal component analysis, established the classification processing model, and applied it to the spectral database systems.This article mainly includes four aspects:Firstly, in the data preprocessing and feature extraction, according to the characteristics of hyperspectral imaging, this paper analysis the spectral characteristics of rock and vegetation, did the researches on the importance of preprocessing. Base on the preprocessing, discussed the feature extraction model for rock and vegetation respectively.Secondly, extract spectrum features on the basis of model, statistical and analysis the result of feature extraction, qualitatively analysis the impact of feature parameters on classification algorithm, extract and analysis rock and vegetation features as the research ways, the results provide data support to feature classification.Thirdly, in hyperspectral data Classification, combined with the feature extraction above, research the classification methods in two major areas:on the one hand is Decision tree classification method base on the spectral dimension, on the other hand is the principal component analysis method for rock. Decision tree classification uses the eight features extracted to do research on classification algorithm based on that spectral features information can reflect the physiological features of vegetation. Principal component analysis method uses the first two principal component to do rock classification based on that principal component can highlight the important features. Experimental results show that using models extract spectrum features, statistical and analysis the result of extraction, combine the conclusion with classification algorithms effectively, will be able to efficiently classify rock and vegetation.Finally, established the classification model by the classification method above and applied to the prototype system of spectral data analysis. The model can be effective in classification and achieve the purpose of analysis and processing spectral data. All of that will lay the foundation for application of hyperspectral remote sensing.