A hyperspectral image classification method based on sparse adaptive semi-supervised multi-manifold learning

Abstract

The invention provides a hyperspectral image classification method for sparse adaptive semi-supervised multi-manifold learning, which proposes a semi-supervised sparse multi-manifold learning dimensionality reduction algorithm and a nearest neighbor multi-manifold classification algorithm. Labeling a small number of data points in the data sample, combined with some unlabeled data points for learning, can well reveal the intrinsic properties and multi-manifold structure hidden in high-dimensional data, and extract low-level data with better discrimination performance. Dimensional embedding features, so as to improve the classification effect and improve the classification accuracy of object categories in hyperspectral remote sensing images, so it can effectively solve the "out-of-sample learning" of sparse manifold clustering and embedding algorithms and the difficulty of labeling categories in remote sensing images problem; at the same time, the experimental results on the PaviaU data set show that, compared with the recognition methods commonly used in the prior art, the method of the present invention has a better classification effect.

Description

Technical field [0001] The invention relates to the technical field of hyperspectral data processing methods and applications, in particular to a hyperspectral image classification method of sparse adaptive semi-supervised multi-manifold learning. Background technique [0002] Hyperspectral remote sensing technology has developed rapidly since the 1980s. Its images record the continuous spectrum of ground objects and contain more information, and have the ability to identify more types of ground objects and to classify objects with higher precision. However, because hyperspectral data consists of a large number of bands to form a high-dimensional feature space, the complexity of most algorithms increases exponentially with the dimensionality, and its processing requires more calculations, and its bands are highly correlated and redundant. , At the same time, there are problems such as high dimensionality, and it is easy to be unable to obtain ideal results due to Hughes phenomeno...

AI Technical Summary

Problems solved by technology

Elhamifar et al. proposed a Sparse Manifold Clustering and Embedding (SMCE) algorithm, which can adaptively select data from the same manifold, and these data points from the same manifold span the same Low-dimensional affine subspace, the similarity graph constructed on this basis can better reveal the intrinsic characteristics of different manifolds in the data, and has a good effect in data clustering, but this method is only defined in the training samples , new samples cannot be directly obtained, and there is a problem of "out-of-sample learning", and this method does not make full use of the category information of the training samples, which limits the identification ability of the algorithm

Images