The Research On The Band Selection Methods Of Hyperspectral Remote Sensing Image

The hyperspectral remote sensing image has the characteristics of high spectral resolution,many bands and narrow band width etc.,and these features lead to the high data dimension and large data volume of the hyperspectral image,which cause a lot of information redundancy.Some bands also contain a lot of noise.The above facts have a certain degree of impact on the accuracy and efficiency of data processing.The band selection can keep the physical meaning of the original band and reduce the dimension of data and the amount of data at the same time.It is often used for preprocessing of the hyperspectral image.Therefore,two new band selection algorithms for hyperspectral image in this paper are proposed aiming at the characteristics of the hyperspectral image and the existing problems.One is based on the linear representation of the hyperspectral image band selection algorithm.Firstly,it establishes a linear relationship for a band with the other bands;then the most relevant band is removed as a redundant band which is determined based on the multiple correlation coefficient;finally,the set of minimum bands is obtained by repeating the above process for the remaining bands.In addition,the algorithm is further optimized to make the optimized algorithm more time-saving and improve the computational efficiency greatly.The other is based on the band index of the hyperspectral image band selection algorithm.Firstly,it uses wavelet transform to deal with the noise of the hyperspectral image date;then bands are divided into groups by using the joint skewness and kurtosis figure;finally,the relatively smaller index of the band is removed as a redundant band which is determined based on the size of the band index.The set of minimum bands is obtained in this way.The algorithm makes full use of the advantages of the wavelet transform,the joint skewness and kurtosis figure and the band index,and combines the three methods to select the bands effectively in order to get a better subset of the bands.Two groups of experiments are designed in order to verify the feasibility of the proposed algorithms in this paper.The experimental results show that the proposed algorithms in the paper are feasible and effective in the band selection process,and they also reduce the correlation and redundancy between the selected bands.They provide the theoretical support to reduce the dimension of hyperspectral image.