Unsupervised Robust Projection Learning by Low-Rank and Sparse Decomposition for Hyperspectral Feature Extraction

Xin Song,Heng-Chao Li,Lei Pan,Yang-Jun Deng,Pu Zhang,Li You,Qian Du

DOI: https://doi.org/10.1109/lgrs.2021.3065252

IF: 5.343

2022-01-01

IEEE Geoscience and Remote Sensing Letters

Abstract:Owing to the strong correlation between the spectral bands of hyperspectral images (HSIs), many feature extraction (FE) methods have been proposed to reduce the redundancy of hyperspectral data. However, Euclidean distance-based FE methods are sensitive to noise. To address this issue, this letter proposed a new unsupervised FE method called robust projection learning (RPL) by integrating the low-rank and sparse decomposition with projection learning. Specifically, in order to enhance the discrimination of traditional robust principal component analysis (RPCA), discriminative RPCA (DRPCA) is first proposed by decomposing the raw data into a low-rank part, a discriminative sparse part, and a structured noise. Moreover, for the purpose of redundancy reduction, projection learning is integrated into DRPCA to obtain a projection matrix with robustness and discrimination. To verify the validity of RPL, two real hyperspectral data sets are used for basic comparison and robust analysis. The corresponding experimental results demonstrate that RPL outperforms the comparative FE methods.

imaging science & photographic technology,remote sensing,engineering, electrical & electronic,geochemistry & geophysics

What problem does this paper attempt to address?

The paper primarily addresses the issues of data redundancy and noise sensitivity in Hyperspectral Images (HSIs). HSIs, with their hundreds of bands containing rich spectral information and useful spatial structures, are widely used in fields such as geological exploration, environmental monitoring, and precision agriculture. However, these images are susceptible to variations in illumination, noise, and sensor accuracy, which increases the difficulty of classification and reduces the accuracy of ground object recognition. Additionally, the high dimensionality of the bands brings significant storage and computational costs. To address these issues, the paper proposes a new method called **Robust Projection Learning (RPL)**. This method combines low-rank decomposition and sparse decomposition and integrates them with projection learning techniques to reduce data redundancy and improve robustness to noise. Specifically, the paper first introduces **Discriminative Robust Principal Component Analysis (DRPCA)** to enhance the discriminative capability of traditional Robust Principal Component Analysis (RPCA). DRPCA decomposes the original data into a low-rank part, a discriminative sparse part, and structured noise. Then, to further reduce redundancy, RPL integrates projection learning into DRPCA to learn a projection matrix that maintains robustness and discriminative power. Experiments conducted on two real hyperspectral datasets demonstrate that the RPL method outperforms other feature extraction methods. The experiments include basic comparisons and robustness analysis, showing that RPL not only effectively reduces data redundancy but also resists noise interference, especially when dealing with different amounts of noisy bands. Therefore, this method provides an effective solution for feature extraction in hyperspectral images.

Unsupervised Robust Projection Learning by Low-Rank and Sparse Decomposition for Hyperspectral Feature Extraction

Low-Rank Discriminative Embedding Regression for Robust Feature Extraction of Hyperspectral Images via Weighted Schatten p-Norm Minimization

Feature Extraction for Hyperspectral Images Using Low-Rank Representation with Neighborhood Preserving Regularization

Exploiting the Sparse Characteristics in Probabilistic Feature Space for Hyperspectral Anomaly Detection

Unsupervised Feature Extraction for Hyperspectral Images Using Combined Low Rank Representation and Locally Linear Embedding

Correntropy-Based Spatial-Spectral Robust Sparsity-Regularized Hyperspectral Unmixing

Unsupervised feature extraction by low-rank and sparsity preserving embedding

Hyperspectral Unmixing Using Robust Deep Nonnegative Matrix Factorization

Learning Latent Low-Rank and Sparse Embedding for Robust Image Feature Extraction

Approximate Low-Rank Projection Learning for Feature Extraction.

Enhanced Tensor Low-Rank Representation Learning for Hyperspectral Anomaly Detection

Residual Deep PCA-based Feature Extraction for Hyperspectral Image Classification

Spectral–Spatial Robust Nonnegative Matrix Factorization for Hyperspectral Unmixing

An Efficient Method of Hyperspectral Image Dimension Reduction Based on Low Rank Representation and Locally Linear Embedding

Hyperspectral Feature Extraction Using Sparse and Smooth Low-Rank Analysis

Projection Subspace Based Low-Rank Representation for Sparse Hyperspectral Unmixing

A Sparse and Low-Rank Near-Isometric Linear Embedding Method for Feature Extraction in Hyperspectral Imagery Classification

Spectral–Spatial and Superpixelwise Unsupervised Linear Discriminant Analysis for Feature Extraction and Classification of Hyperspectral Images

Spectral-Spatial Discriminant Feature Learning For Hyperspectral Image Classification

An Efficient and Adaptive Reconstructive Homogeneous Block-Based Local Tensor Robust PCA for Feature Extraction of Hyperspectral Images

Low-Rank Projection Learning Via Graph Embedding