Hyperspectral image suspicious target detection method based on low-rank sparse representation

Abstract

The invention discloses a hyperspectral image suspicious target detection method based on low-rank sparse representation. The method comprises the following steps: reading a hyperspectral image data cube and related projection parameters; clustering spectrums of all pixels in the input hyperspectral image to obtain a plurality of clustering centers and a category to which each pixel belongs; counting an intra-class clustering result, and determining the distance between each pixel and the clustering center of the class to which the pixel belongs; selecting s pixels with the minimum distance asvarious description sample points, and adding the description sample points into a final dictionary to form a reconstruction dictionary of the scene remote sensing data; performing low-rank sparse constraint matrix decomposition on the hyperspectral image data by using the reconstructed dictionary; carrying out residual error statistics on the decomposed residual error; and outputting the T pixels with the maximum residual values and the ground coordinates corresponding to the T pixels as suspicious targets. The detection method provided by the invention has good adaptability to hyperspectralimages collected by different sensors, has a low false alarm rate, and still has a high detection capability for small targets.

Description

technical field [0001] The invention relates to a remote sensing image target detection method, in particular to a hyperspectral image suspicious target detection method based on low-rank sparse expression, and belongs to the technical field of digital image processing. Background technique [0002] Hyperspectral image (HSI) is a spectral band image ranging from visible light to short-wave infrared (0.4-2.5 microns), sampled with a spectral resolution of a few nanometers, and in the order of tens to hundreds of bands. It uses narrow-band imaging data technology and spectral technology to detect the two-dimensional geometric space and one-dimensional spectral information of the target, and obtain continuous and narrow-band image information with spectral resolution. Hyperspectral images combine image information and spectral information of samples. Since different components have different spectral absorptions, the image will reflect a certain defect more significantly at a ...

AI Technical Summary

Problems solved by technology

However, this method is not effective in modeli

Images