Research Of Feature Extraction/Selection Techniques And Application Based On Hyperspectral Image

The Hyperspectral image has the character of combining the image and spectrum, it is a new remote sensing technique recently developed. Compared to Multi-spectral image, hyperspectral image is high spectral resolution, narrow band, and has many bands. It can distinguish targets with reliability. However, these advantages are at the expense of high dimension and large data amount, besides, the high relativity of hyperspectral image bands makes much Information redundancy. And there is no need to use every band to do classification and targets identify, so it is necessary to do dimensionality reduction first. This paper researches dimensionality reduction both from feature extraction and selection techniques.This paper first introduce basic conception and physics elements about hyperspectral image, at the same time introduce the basic conception of feature extraction and selection techniques and the research status in quo, and classification, target detection, then analysis character of hyperspectral image, including its data expression and three characters.We study on the image information and quantitative analysis focusing on the spectral dimension because the hyperspectral data has the abundant spectral information. Based on the first two chapters, this paper researches feature extraction both from wavelet transform and spectral absorption features. The wavelet transform theory has been developed furiously in the decades in the field of image processing. In analysis of time field, wavelet transform is a new fast developed implement, has good localize character both in time and frequency field, its a relatively perfect method of spectral feature extraction. This paper researches one-dimension high frequency decompose, use the gained local characters to identify targets. Absorption-band parameters such as the position, depth, width, and asymmetry of the feature have been used to quantitatively estimate composition of samples from hyperspectral data. So spectral absorptions are very important feature bands in use of hyperspectral classification and targets identify, this paper extracts absorption features of actual hyperspectral image by continuum removed method which is very useful. Then tested both of them by...