Feature Extraction And Matching Method Research Of The Core Of Hyperspectral Based On Wavelet Transform

The essence of feature extraction is to extract the characteristics which can describe the model best from a large number of original data. It’s an effectivepretreatment method to overcome the "Curse of dimensionality" problem. Spectral feature extraction is to extract the useful and can distinguish with other data informationeffectively from the original spectral data, so as to realize the spectral recognition.Previously, people can only use the Fourier transform to research and analysis the characteristics of the signal in frequency domain or time domain, but can’t get the frequency characteristics of one specific timenearby. Because the core ofhyperspectral signal has a variety of characteristics, there isa lot of limitations to deal with signalby Fourier transform. Recent years, wavelet transform has become more and more mature.It makes the signal analysis achieve the time-frequency localization. To decompose and reconstruct with signal by wavelet, and then process and process information in the frequency domain, has become an important way from feature extraction. Wavelet transform can better extract the useful characteristics from the spectral data,that is to say, wavelet method has become animportant and effective mathematical tool for spectral data analysis and practical application.The main task of this paper is to use wavelet transform technique for hyperspectral core’sprocessing,the purposeis to extract the useful information features. This article is intended to applythe wavelet analysis of hyperspectral core feature extraction, in order to find a method to process and analyze the data easily and to ensure that this approach is feasible. This paper chooses Jiguanzui copper-gold mine as the study area, and combine with the actual situation in the study area.By wavelet decomposition, select the appropriate wavelet coefficients as the characteristics of the mineral spectral matching.This paper uses three wavelet methodsto describe the features: the mean of low frequency and high frequency coefficients, and each scale’s energy value of wavelet coefficients. Meanwhile, this paper uses waveform feature extraction methods based on the study area,the purpose is to extract the core hyperspectral featureson study area.Spectral matching is based on the similarity of the two spectral curves to achieve the purpose of identifying the ownership of the feature categories.The conventional spectral matching methodsare not applicable for alarge number of data of the original spectrum. Because when reflectance data is too much,thetime computation complexity and space complexity will increasegreatly. At this time, we will try to extractfeaturesof core hyperspectral of Jiguanzui copper-gold mine.The featuresmeansthe features extractedby wavelet method,and then do pattern recognition for these feature vectors. This papermainly uses two methods tomatch the mineral feature vectors: the angle between the two feature vectorsand the Euclidean distanceof the two feature vectors. Finally,weget to know whether the matching result is good or bad, and of course we can judgewhich way to describewavelet features is better.In a word, this paper selects wavelet transform to deal with the core of hyperspectral data, in order to extract its featureand then match the characteristic of mineral spectrum. Therefore, it can further the analysis of the features, so as to identify the type of mineral.