Target Feature Extraction Of Hyperspectral Remote Sensing Target Based On Mask

During the last 20 years, hyperspectral remote sensing has been playing an importantrole in many fields of both military and civil applications。Compared Multi-spectrum remote sensing,the characteristics of hyperspectral remote sensing data are more of channels, higher spectral resolution, narrower bandwidth and larger amount of data。This for utilize remote sensing image carry on goal classify,discern and with importance value of research of following etc. But its enormous data amount and higher data bring greater difficulty to classifying of hyperspectral remote sensing image。The special properties of hyperspectral image data are first analyzed,Prove hyperspectral remote sensing image relatively stronger spectrum dependence, and analyzed its challenging influence on traditional classification methods 。 Before classification with usual classification method, feature selection and abstract should be carried through. Transform based on mathematical method with linear and nonlinear transform, linear transformations are commonly used principal component analysis(PCA), the most small autocorrelation factor method(MAF), minimum noise fraction(MNF), projection pursuit method etc.. The linear transformation of the original information of multiple bands in be concentrated on a number of new groups of as few as possible points in the image, the image information merging processing, the image data quantity is effectively compressed, uncorrelated between the components but also can make new components in the image. The principal component transform first of all the original band data in hyperspectral images obtained by correlation coefficient matrix between images. Secondly, from the correlation coefficient matrix to calculate eigenvalues, and sorted in order from large to small, the feature vector corresponding to the characteristic equation. Finally, through a preset threshold(usually a value of 95%) to a former principal component. In this paper the study area hyperspectral image processing PCA transform, the experimental data shows that the effective information content contained four principal components has reached more than 95%.Hyperspectral remote sensing image to obtain the image data and the narrow spectrum through continuous imaging spectrometer, for each pixel provides spectral information of narrow band of tens to hundreds of, and generate a complete and continuous spectrum curve, the spectral characteristics of the fine expression. The spectral curve reflects the absorption features, and spectral curves of different substances are also different. Based on this, can extract features from the spectral curve. Based on this, first introduced the spectral absorption feature parameters based on the commonly used, absorption depth, width of spectral absorption parameters, continuum removal, spectral absorption characteristics index, spectral coding, spectral derivative and other commonly used extraction algorithm. The characteristics of original spectral curve analysis of area objects. The analysis of the Meng Tuoshi and kaolinite spectral curve in this paper, compared to the same material(Meng Tuoshi) spectral absorption characteristics, the spectral characteristic which can represent the type of material. Compared with the same method of different material(montmorillonite and kaolinite) spectral absorption characteristics, find out which can distinguish the effective features of these two kinds of material. In addition to the two kinds of material spectral curve derivation, spectral analysis of derivation after absorption characteristics, get the same substance and different substances effectively identify characteristics. The test shows that the higher order derivative spectra, to better.According to the characteristics of spectral parameters mentioned above and absorption spectra extraction method, proposed a feature extraction algorithm based on mask. In the proposed algorithm, this paper introduces the basic principle of the mask, and method of making mask. By two spectral curve comparison algorithm(spectral curve of standard spectral library and curve to identify the spectral similarity(Similarity). Of course, before the need for two spectral curves of three order derivative processing, because such treatment would make spectral features more prominent, easy to identify. If Similarity and 1 of the difference in the allowed range, the image element is like target element, code 1, code 0 otherwise. This process is a process of mask.Experimental results show that the image obtained by this algorithm in the feature extraction of the clear, is an effective feature extraction algorithm.