Hyperspectral image dimension reduction method based on clustering multiple manifold measure learning

Abstract

The invention provides a hyperspectral image dimension reduction method based on clustering multiple manifold measure learning, and manifold learning is introduced on the basis of multiple local measure learning. Original complex non-linear data is decomposed into a plurality of correlative data through multi-measure learning, so that details in a data structure can be better grasped in a training process, over-fitting is avoided to a certain extent while targeted training is performed, and the non-linear data can be well processed. On the above basis, while data label information is utilized, structural information of the data can be effectively utilized, and the limited prior information is utilized to the maximum extent. While dimension reduction is carried out, the finally obtained feature space can effectively distinguish different classes, which is helpful for subsequent classification of images. The method has the advantages that both linear data and linear data can be processed, meanwhile, overfitting in classification calculation is avoided, and classification precision is improved.

Description

technical field [0001] The invention relates to the technical field of remote sensing image processing, in particular to a hyperspectral image dimensionality reduction method based on clustering multiple manifold measure learning. Background technique [0002] Hyperspectral remote sensing image processing plays an important role in information detection and is an important topic in the field of remote sensing. Hyperspectral images contain hundreds of spectral bands, providing detailed spatial structure and spectral information. Therefore, the subtle differences between land cover categories can be distinguished through hyperspectral images, which are widely used in atmospheric monitoring, vegetation monitoring, disaster warning and other fields. With the continuous development of sensors, the dimensionality of data is also increasing. However, the increase in data dimensionality does not improve the classification and recognition capabilities of images for ground objects, b...

AI Technical Summary

Problems solved by technology

However, both the global measure learning method and the multi-local measure learning method ignore the structural characteristics of the data, and cannot use the structural information of the data itself.

This means that the algorithm does not take full advantage of the information provided by the training samples

However, the acquisition cost of prior samples in hyperspectral data is very expensive, so how to maximize the use of limited sample information is very important.

Images