8 February 2024 in IEEE Transactions on Geoscience and Remote Sensing
Deep learning based hyperspectral image (HSI) change detection (CD) has been a research hotspot. However, the high dimensionality and the limited training samples make HSI-CD difficult to implement. Besides, the data uncertainty inherited from HSIs is often neglected. To deal with these issues, this paper presents an uncertainty aware graph self-supervised learning (UA-GSSL) approach for unsupervised HSI-CD, which allows encoding not only the spectral and topological structure attributes but also the data uncertainty into learned feature representation for downstream CD task. Specifically, spectral and spatial correlations in HSIs are firstly characterized via graph model. Then, based on the constructed spectral-spatial graph models, novel node-level and edge-level data augmentations are devised to enrich the contrastive sample pairs. Thirdly, a graph based dual-branch self-supervised learning (SSL) contrastive network is designed to maximize the mutual information between a pair of low-dimensional feature embeddings. Fourthly, a simple but effective uncertainty aware loss function is dedicatedly formed to encourage the reliable features to play the more dominant role in feature representation. Finally, change map is produced using the learned features. Experimental results obtained on four real HSI datasets sufficiently demonstrate that the UA-GSSL achieves remarkable results compared to twelve state-of-the-art (SOTA) methods. The source code of this article can be downloaded from https://github.com/vickyiiiii /UA-GSSL.
18 October 2024 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Deep learning has shown promising performance in the field of hyperspectral image (HSI) change detection (CD). However, most of these methods often focus on local spatial-spectral information but ignore the high-order correlations contained in multitemporal HSIs. To address this issue, this article proposes a hypergraph self-supervised learning (HG-SSL) based joint spectral-spatial-temporal feature representation algorithm (HyperSST) for downstream HSI-CD. Inspired by the process of human brain perception, HyperSST uniformly models the spectral-spatial-temporal correlations in the form of high-order interactions and skillfully exploits the vertex-level, hyperedge-level, and vertex/hyperedge-level inherent structures within the unlabeled multitemporal HSIs. Specifically, two types of hyperedges, spectral-spatial correlation hyperedge and temporal correlation hyperedge, are first formulated to fully exploit the high-order spectral, spatial, and temporal interactions contained in bitemporal images. Second, two SSL strategies namely contrastive spectral-spatial features learning and generative temporal features learning are skillfully designed to exploit the inherent characteristics of hypergraph models and extract discriminative latent feature representations for downstream tasks. The former captures changes in both attribute space and structure space, while the latter apprehends the changes in temporal space. Third, the learned joint spectral-spatial-temporal features provide a comprehensive representation to qualify the changes between multitemporal images. Extensive experiments on four challenging HSI datasets demonstrate the effectiveness of the proposed approach.
23 September 2025 in Expert Systems with Applications