Hyperspectral image classification method based on nuclear low-rank representing graph and spatial constraint

Abstract

The invention belongs to the technical field of image processing, and particularly provides a hyperspectral image classification method based on a nuclear low-rank representing graph and spatial constraint. The method includes the implementing steps that (1), samples of all known labels of a hyperspectral image serve as training samples, samples of unknown labels serve as test samples, and a sample set is constructed according to the sequence that the training samples are located in front of the test samples; (2), column normalization is conducted on the sample set and nuclear space mapping is conducted; (3), low-rank representation is conducted on the sample set obtained after nuclear space mapping, so that the nuclear low-rank representing graph is acquired; (4), a space information graph of the sample set is established; (5), the low-rank representing graph and the space information graph are summed, so that a new graph is established; (6), according to a graph maintaining standard method, category labels of the test samples are acquired. The method mainly overcomes the defect that the classification accuracy is low when training samples are insufficient in an existing method, meanwhile, the features of the hyperspectral image are reasonably considered, a better classification result can be acquired in the combination of space information, and the better robustness and the higher accuracy are achieved by the adoption of a nuclear low-rank method.

Description

technical field [0001] The invention belongs to the technical field of image processing, and relates to a method for classifying hyperspectral images of semi-supervised images, which can be used to complete the classification of hyperspectral images in the case of a small number of labeled samples, specifically a hyperspectral image based on kernel low-rank representation graphs and space constraints Image classification methods. Background technique [0002] Imaging Spectroscopy is a new research field developed in the field of remote sensing based on Spectroscopy in the 1980s. The spectral resolution of the sensor is 10 -2 The remote sensing of lambda is called hyperspectral remote sensing, and it has tens to hundreds of bands in the visible light to near-infrared spectral region. Hyperspectral remote sensing can use many narrow electromagnetic wave bands to obtain relevant data from objects of interest, organically integrate image dimension information and spectral dimen...

AI Technical Summary

Problems solved by technology

[0005] For the classification and prediction of ground objects in the practical application of hyperspectral remote sensing images, the current supervised method is better. The supervised method needs a large number of correct training samples, that is, it needs to obtain the category labels of the ground objects. However, in many cases, the ground objects Marking is difficult to obtain, and field surveys take a lot of manpower and material resources. In many emergency situations (such as landslides, forest fires, earthquakes, floods), it is impossible to obtain ground feature marks

Images