Hyper-spectral image classification method based on local manifold embedding

Abstract

The invention discloses a hyper-spectral image classification method based on local manifold embedding. The hyper-spectral image classification method includes the steps: 1) reconstructing data points of a training sample by the aid of similar neighbor points; 2) constructing an intra-class graph, an intra-class reconstruction graph, an inter-class graph and an inter-class reconstruction graph by the aid of neighborhood of the data points and reconstruction points corresponding to neighborhood points; 3) keeping structures of the intra-class graph and the intra-class reconstruction graph unchanged in a low-dimensional embedding space, and inhibiting the structural relation between the inter-class graph and the inter-class reconstruction graph to obtain a projection matrix from a high-dimensional space to the low-dimensional space; 4) acquiring low-dimensional embedding characteristics of the training sample; 5) reducing dimensionality of high-dimensional data of a testing sample by the projection matrix to obtain testing sample low-dimensional embedding; 6) classifying testing sample low-dimensional embedding by a classifier to obtain hyper-spectral image classification results. Internal implication attributes of hyper-spectral images can be more effectively represented, diagnostic characteristics can be more effectively extracted, and data divisibility is improved.

Description

technical field [0001] The invention relates to hyperspectral image classification, in particular to a hyperspectral image classification method based on local manifold embedding, and belongs to the technical field of hyperspectral image classification. Background technique [0002] Scientific researchers proposed hyperspectral remote sensing based on multispectral remote sensing in the early 1980s. The spectral resolution of hyperspectral remote sensing images is as high as 10 -2 λ order of magnitude (belonging to the nanoscale), the band ranges from visible light to short-wave infrared, and the number of spectral bands is as many as dozens or even hundreds. The high spectral resolution of hyperspectral image data makes the interval between adjacent bands narrow , there is an overlapping region of bands, and the spectral channels are no longer discrete but continuous, so hyperspectral remote sensing is often called imaging spectral remote sensing. Hyperspectral remote sen...

AI Technical Summary

Problems solved by technology

[0044] Aiming at the deficiency that MFA cannot effectively represent the internal manifold of data, the purpose of the present invention is to provide a method that can better characterize the intrinsic properties of hyperspectral images, extract discriminative features more effectively, and improve data separability. Hyperspectral Image Classification Method Based on Local Manifold Embedding

Images