A Sparse Representation Method of Imaging Spectral Image Based on Redundant Dictionary of Object Category Classification

Abstract

The invention discloses a method for sparse representation of imaging spectral images based on redundant dictionary of classification of ground objects, including K-means, training of classification redundant dictionary based on K-SVD, and full-band sparse representation of imaging spectral images. Through unsupervised K-means clustering, the information of the imaging spectral image is divided into corresponding subsets according to the object category, and the imaging spectral image is sparsely decomposed through the classification dictionary. On the premise of ensuring the sparsity, the information of all bands of the reconstructed image is improved. quality. It makes full use of the characteristics of similar spectral curves of the same object category in the spectral image, avoids the introduction of wrong information, ensures the quality of the reconstructed spectral image, and effectively avoids the problem of spectral distortion caused by independent processing of each band.

Description

technical field [0001] The invention relates to a sparse representation method of an imaging spectral image, in particular to a sparse representation method of an imaging spectral image based on a redundant dictionary of classification of ground objects. Background technique [0002] Imaging spectroscopy technology is a new remote sensing technology that developed rapidly in the 1980s. It obtains the radiation information of surface features in dozens or even hundreds of bands at the same time with ultra-high spectral resolution at the nanometer scale, so it can acquire ground features While collecting the spatial information of the target, it detects its internal physical characteristics and composition. Imaging spectroscopy technology is widely used in resource exploration, geological exploration, disaster relief and other fields. [0003] Similar surface areas have similar reflectance values ​​and therefore similar spectral curves. The imaging spectral image can be rega...

AI Technical Summary

Problems solved by technology

The basis tracking algorithm is extremely complex because it needs to minimize a global objective function in all dictionary vectors

[0007] The sparse representation method of the imaging spectral image in the existing technology does not fully consider the ability to maintain the spectral dimension information in the imaging spectral image, resulting in a certain spectral distortion in the high-resolution image reconstructed by the sparse representation signal

In addition, there are many bands in hyperspectral images, and it is often unbearable to perform sparse decomposition on all bands.

Images