[论文解读] SLCGC: A lightweight Self-supervised Low-pass Contrastive Graph Clustering Network for Hyperspectral Images

Yao Ding, Zhili Zhang+5 more|ArXiv.org|Feb 5, 2025

Remote-Sensing Image Classification被引用 8

一句话总结

A self-supervised, lightweight hyperspectral image clustering framework that uses region-based graph construction, low-pass denoising, and dual-branch contrastive learning to produce robust, efficient embeddings for K-means clustering.

ABSTRACT

Self-supervised hyperspectral image (HSI) clustering remains a fundamental yet challenging task due to the absence of labeled data and the inherent complexity of spatial-spectral interactions. While recent advancements have explored innovative approaches, existing methods face critical limitations in clustering accuracy, feature discriminability, computational efficiency, and robustness to noise, hindering their practical deployment. In this paper, a self-supervised efficient low-pass contrastive graph clustering (SLCGC) is introduced for HSIs. Our approach begins with homogeneous region generation, which aggregates pixels into spectrally consistent regions to preserve local spatial-spectral coherence while drastically reducing graph complexity. We then construct a structural graph using an adjacency matrix A and introduce a low-pass graph denoising mechanism to suppress high-frequency noise in the graph topology, ensuring stable feature propagation. A dual-branch graph contrastive learning module is developed, where Gaussian noise perturbations generate augmented views through two multilayer perceptrons (MLPs), and a cross-view contrastive loss enforces structural consistency between views to learn noise-invariant representations. Finally, latent embeddings optimized by this process are clustered via K-means. Extensive experiments and repeated comparative analysis have verified that our SLCGC contains high clustering accuracy, low computational complexity, and strong robustness. The code source will be available at https://github.com/DY-HYX.

研究动机与目标

• Address the challenge of clustering hyperspectral images without labeled data.
• Improve clustering accuracy while reducing computational complexity.
• Enhance robustness to noise and unwanted high-frequency graph components.
• Preserve local spatial-spectral coherence during clustering.
• Provide an end-to-end pipeline that enables efficient K-means clustering on learned embeddings.

提出的方法

• Generate homogeneous regions to aggregate spectrally consistent pixels and reduce graph size.
• Construct a structural graph with an adjacency matrix A.
• Apply a low-pass graph denoising mechanism to suppress high-frequency noise in the graph topology.
• Implement a dual-branch graph contrastive learning module with Gaussian noise perturbations and two MLPs to create augmented views.
• Use a cross-view contrastive loss to enforce structural consistency and learn noise-invariant representations.
• Cluster the resulting latent embeddings with K-means.

实验结果

研究问题

• RQ1 How effective is homogeneous region generation for preserving spatial-spectral coherence while reducing graph complexity?
• RQ2 Does low-pass graph denoising improve robustness and stability of feature propagation on HSIs?
• RQ3 Can dual-branch contrastive learning with cross-view loss produce noise-invariant representations for clustering?
• RQ4 What are the improvements in clustering accuracy and computational efficiency compared to baseline methods?

主要发现

• Demonstrates high clustering accuracy on hyperspectral data.
• Exhibits low computational complexity relative to comparable self-supervised methods.
• Shows strong robustness to noise through low-pass denoising and contrastive learning.
• Provides stable feature propagation via the proposed graph structure.
• Learns embeddings that facilitate effective K-means clustering.