GF-5 Hyperspectral Data Preprocessing And Analysis Of Feature Extraction Methods

GF-5 is the latest hyperspectral comprehensive observation satellite launched in China.And the application research of GF5 is also a hot spot in the field of remote sensing at home and abroad.In this paper,an effective preprocessing method for hyperspectral remote sensing data of GF-5 is tried.In order to achieve better application effect,data fusion and feature extraction are carried out.Traditional spectral matching method,support vector machine classification method and ENVINet5 deep learning method are selected to study of feature extraction and classification,and the main results are as follows:(1)In this paper,ENVI were used to preprocess the hyperspectral data of GF-5.Through a series of processing,the radiation and geometric distortion of the image were corrected,and the true reflectance of the ground object was restored.The result is good.(2)After the fusion of gf-5 hyperspectral data(30M)and Landsat 8 panchromatic band(15m)by GS method,the texture details of the image are improved and the spectral information is well preserved.The feature extraction of hyperspectral data was carried out by block principal component analysis.And the band correlation is reduced.(3)For the hyperspectral data of GF-5,six kinds of features in suburbs were classified and extracted by using spectral angle matching(SAM),support vector machine(SVM)and ENVINet5.The kernel function selection of SVM and the weight parameter adjustment of envinet5 were tested respectively.The classification results of the three methods were analyzed comprehensively.The overall accuracy is more than 80%,which shows that the GF-5 hyperspectral data has a good effect in the application of ground feature extraction and classification,and has a good application prospect in fine classification.SAM is sensitive to the spectral curve,which is suitable for similar object extraction with difference.All kinds of results of SVM are stable,and it is suitable for large area and wide distribution of ground objects.The precision of envinet5 is the highest,especially for the linear regular feature extraction.