Method and device for selecting hyperspectral image band based on key band extraction

Abstract

The invention discloses a method and device for selecting a hyperspectral image band based on key band extraction. The method is specifically implemented by: determining the number of end members of a hyperspectral image, and extracting end member spectrums; extracting a key point subset for each end member spectrum by using a method based on three-point vector included angle and variation amplitude analysis, and combining all the key point subsets to construct a candidate band subset; according to the characteristic that inter-band similarities have clustered distribution, constructing a visibility graph of a local information divergence matrix, and determining a range of an optimum band number; and finally, selecting the best one of an information quantity measuring method and an optimum subset selection criteria method, thereby determining the optimum band subset. Key bands of the end member spectrums provided in the method disclosed by the invention are most distinguishing feature bands among different ground objects, an optimum band is selected from the key band subset, and the time of subsequent band selection can be shortened, so that the band selection method disclosed by the invention can improve the efficiency of band selection.

Description

technical field

[0001] The invention relates to the field of image processing, in particular to a hyperspectral image band selection method and device based on key band extraction. Background technique

[0002] Hyperspectral remote sensing is a new type of remote sensing detection technology developed in recent years, which has broad application prospects. Hyperspectral remote sensing images generally consist of hundreds of bands and contain rich spatial, radiometric and spectral information. However, a large number of bands increases the time for hyperspectral image anomaly detection, and the correlation between bands reduces the detection accuracy. Therefore, the prerequisite for effective use of hyperspectral data is to select appropriate features to reduce the dimensionality of hyperspectral data. There are two existing methods to achieve dimensionality reduction: one is feature extraction, and the other is band selection. The method of feature extraction will cause th...

Images