The surge in remote sensing (RS) data underscores the need for improved data diversity and processing. While integrating hyperspectral (HS) and LiDAR data enhances analysis and addresses spectral variability, the high dimensionality, noise, and outliers inherent in hyperspectral images present significant challenges. Additionally, the precise labeling required for HS makes supervised classification labor-intensive, professional-focused, and time-consuming, further motivating the development of advanced HS clustering algorithms to address these issues. Unsupervised clustering addresses the above issues but still struggles due to underutilization of auxiliary spatial and structural information, high data dimensionality with redundant hyperspectral bands, and information divergence from heterogeneity among multimodal data. These challenges impede the effective extraction of consistent structures, undermining clustering stability and overall model performance. To address these challenges, we propose a Superpixel-Based Bipartite Graph Clustering Enriched with Spatial Information for Hyperspectral and LiDAR Data (SBGC) model. Our proposed method fully utilizes spatial information to construct meaningful bipartite graphs for the efficient processing of multimodal remote sensing data. By adopting a projected clustering paradigm, our approach simultaneously clusters and reduces dimensionality, effectively eliminating redundant bands. Additionally, it innovatively stacks multimodal data into tensors, thoroughly exploring the consistent structures in the low-rank space among different modalities. This reduces the heterogeneity-induced information divergence and significantly enhances clustering performance. Extensive experiments on real datasets confirm the method’s effectiveness and advanced capabilities.
6 November 2024 in IEEE Transactions on Circuits and Systems for Video Technology
Ensemble clustering based on co-association matrices integrates multiple connective matrices from base clusterings to achieve superior results. However, these methods primarily focus on inter-sample relationships, neglecting variations across different base clusterings, potentially introducing noise. Additionally, they overlook interactions between samples and base clusterings, which are crucial for extracting common information and avoiding post-processing steps that may cause information loss and instability in clustering results. To address these issues, we propose the Tensorized Graph Learning for Spectral Ensemble Clustering (TGLSEC) model. TGLSEC stacks all connective matrices into a third-order tensor, employs Fast Fourier Transform (FFT) for encoding, and elucidates interrelations in the frequency domain. By minimizing the tensor Schatten p -norm, TGLSEC extracts common information in the low-rank space, eliminating noise and improving the quality of the common shared graph. Incorporating Laplacian rank constraints, TGLSEC learns a common shared graph with c-connected components, directly representing the clustering structure and avoiding post-processing steps, leading to more stable clustering results. To enhance computational efficiency for large-scale datasets, TGLSEC has been expanded into a bipartite-graph-based model, TGLSEC-BG, reducing complexity and computational time. Extensive experiments on real-world datasets demonstrate that TGLSEC and TGLSEC-BG exhibit superior clustering performance and robustness to noise.
Unsupervised hyperspectral image (HSI) clustering is a fundamental yet challenging task due to high dimensionality and complex spectral–spatial characteristics. In this paper, we propose a novel and efficient clustering framework centered on adaptive and diverse anchor graph modeling. First, we introduce a parameter-free construction strategy that employs Entropy Rate Superpixel (ERS) segmentation to generate multiple anchor graphs of varying sizes from a single HSI, overcoming the limitation of fixed anchor quantities and enhancing structural expressiveness. Second, we propose an anchor-to-pixel label propagation mechanism to transfer anchor-level cluster labels back to the pixel level, reinforcing spatial coherence and spectral discriminability. Third, we perform clustering directly at the anchor level, which substantially reduces computational cost while retaining structure-aware accuracy. Extensive experiments on three benchmark datasets (Trento, Salinas, and Pavia Center) demonstrate the effectiveness and efficiency of our approach.
16 October 2025 in IEEE Transactions on Geoscience and Remote Sensing
The increasing complexity of remote sensing (RS) applications necessitates multimodal data fusion to overcome the inherent limitations of single-source data. In particular, the integration of hyperspectral images (HSI) and LiDAR data captures complementary spectral and spatial features, significantly enhancing object discrimination. Unlike supervised methods requiring costly expert annotations, unsupervised clustering eliminates labeling needs, thereby offering an efficient solution for complex scene analysis with reduced deployment costs. However, existing unsupervised clustering methods still face several challenges, including insufficient utilization of spatial information, high data dimensionality, and interference from high-frequency noise, which collectively hinder their further development and practical application. To address these issues, this paper proposes a novel method called Spatial-Spectral Bipartite Graph Clustering with Low-Frequency Tensor Regularization (SSBC-LFT) for Hyperspectral and LiDAR Data. The proposed method first focuses on modeling clustering structures, innovatively constructing spatially-informed bipartite graphs for each modality to enrich spatial details while reducing computational complexity. Additionally, an adaptive spectral band selection mechanism dynamically assigns higher weights to discriminative bands during clustering while suppressing redundant ones. Finally, through proposed Low-frequency Tensor Nuclear Norm regularization, high-frequency noise is effectively filtered in the frequency domain, extracting stable and consistent cross-modal structural information from the low-rank space to further enhance clustering robustness and accuracy. Extensive experiments on real-world multimodal RS datasets demonstrate the superior performance of SSBC-LFT model, offering a robust and effective solution for unsupervised clustering in complex scenarios.
5 May 2026 in IEEE Transactions on Circuits and Systems for Video Technology
Multimodal remote sensing (RS) clustering synergistically integrates multi-dimensional information, effectively addressing the representational limitations of single-modality data. This integration offers essential technical support for fine-grained recognition and accurate interpretation of ground objects in complex scenarios. However, existing methods still face several challenges, including insufficient utilization of spatial information, limited ability to extract consistent information due to inter-modal heterogeneity, and low efficiency when handling large-scale and complex datasets. To address these issues, we propose an Integrating Pseudo-Supervision and Spatial Constraints for Efficient Clustering (PSSC) of Multimodal Remote Sensing Data model. The proposed method begins by constructing spatial bipartite graphs from multimodal data to fully exploit spatial information while reducing computational complexity. These graphs are then stacked into a third-order tensor, upon which a robust denoised representation is learned to suppress noise and preserve the core structural characteristics of the multimodal inputs. Based on this clean tensor, PSSC captures cross-modal consistency by minimizing the tensor nuclear norm within the low-rank space. To further enhance clustering efficiency and accuracy, a region homogeneity-constrained rapid label generation strategy is proposed, which leverages high-confidence pseudo-supervision information from homogeneous regions to iteratively refine clustering labels, thereby significantly reducing computational overhead. Extensive experiments on real-world multimodal datasets validate the effectiveness and superior performance of the proposed method.
5 June 2026 in IEEE Transactions on Pattern Analysis and Machine Intelligence
Multi-view clustering methods based on graph learning have attracted considerable attention due to their superior clustering performance. However, such approaches generally lack effective supervisory signals, making it difficult to fully exploit the latent correlations among multi-view data. In addition, existing methods often treat all samples equally in local learning, which limits their ability to capture critical local structures. To address these issues, A Unified Framework for Pseudo-Supervised Clustering via Weighted Sample Aggregation (PSC-WSA) is proposed, which constructs a complete pseudo supervision-guided clustering framework encompassing pseudo supervision information generation, weighted sample aggregation, clustering, and label propagation. In this framework, the sample aggregation process is learnable under the constraints of pseudo-supervision, and synergistically interacts with the clustering process, serving as the key bridge for pseudo-supervision guided clustering. Moreover, two weighted aggregation strategies are designed to adaptively model local relationships between highly similar samples. This enables diverse sample-level locality learning across different views, thereby effectively enhancing the discriminability of the representative samples. To optimize the PSC-WSA model, an efficient alternating iterative optimization algorithm is developed. Extensive experiments on both multi view datasets and multimodal remote sensing datasets validate the feasibility, effectiveness and strong scalability.