Superpixel-based graph methods are efficient for hyperspectral image (HSI) classification, yet they often rely on static graph topologies fixed during preprocessing. These predetermined structures, based primarily on spatial adjacency, generally fail to capture the richer semantic correlations present in HSI data. Moreover, superpixel node features are often obtained by simple averaging, making them susceptible to mixed noise and spectral heterogeneity. To overcome these limitations, we propose a novel framework termed GSDG-Net, which integrates geometry-guided dynamic graphs with Fourier-modulated superpixels. Instead of using a fixed adjacency graph, our approach first prunes long-range spatial connections based on centroid distances to obtain a relatively sparse prior graph. It subsequently applies a content-aware attention mechanism to dynamically re-weight the remaining edges, explicitly prioritizing semantic affinity to construct a semantically-aware dynamic graph. To address the issue of feature ambiguity, we devise an innovative Fourier-Domain Spectral Modulation (FDSM) module. A differentiable complex-valued fourier-domain gating mechanism lies at the heart of this module. It acts as an adaptive, learnable spectral filter that emphasizes informative low-frequency components and attenuates high-frequency noise, directly yielding discriminative Fourier-modulated superpixel features. Further augmenting the representation, a multi-scale convolutional branch compensates for the pixel-level detail loss that arises from superpixel abstraction. Experiments on four benchmark HSI datasets demonstrate that GSDG-Net achieves state-of-the-art performance under few-shot settings. Code is available at https://github.com/LanLanZhao/IEEE_TGRS_GSDG-Net.

Jointly clustering multi-source remote sensing (RS) data remains challenging despite the promising progress of multi-view clustering (MVC). Many existing methods have several limitations: 1) two-stage pipelines that separate feature learning from clustering often incur error accumulation and lead to blurred land-cover boundaries; 2) fixed or equal-view fusion cannot adapt to heterogeneous view quality, allowing noisy or redundant views to dominate the consensus representation; and 3) the construction of full affinity matrices results in high computational cost, restricting scalability for large RS data. To address these issues, we propose an Efficient One-step Orthogonal Consensus (EO2C) framework for multi-view RS clustering. EO2C first performs superpixel-guided spatial denoising to obtain reliable and spatially coherent features. It then learns structure-consistent embeddings by jointly optimizing the embedding, mapping matrix, and cluster indicator matrix, thereby preserving view-specific local structures. More importantly, an adaptive consensus fusion strategy based on a collective orthogonality constraint projects reconstructed view-specific representations into a shared consensus space while automatically regulating view contributions to suppress unreliable views. These components collectively enable EO2C to achieve robust, discriminative, and one-step clustering. An efficient alternating optimization algorithm further attains near-linear computational complexity via closed-form or orthogonal updates. Extensive experiments on four real-world multi-source RS datasets demonstrate that EO2C outperforms state-of-the-art methods in clustering performance. The code is available at https://github.com/ytccyw/EO2C.

Multi-view clustering (MVC) typically enhances clustering performance by leveraging consistent and complementary information across views. With increasing data scale, many MVC methods employ few anchors to represent data, effectively reducing complexity. Traditional anchor selection methods include random sampling and k-means. Recent methods leverage variance or inter-class error to guide anchor selection. However, their anchors are selected sequentially, where each newly selected anchor focuses on divergences from the previous anchor. This local perspective leads to unstable anchor coverage and fails to fully represent complex cluster shapes. An anchor selection algorithm, termed DPAS, is proposed to solve this problem. From a novel density perspective, it selects density peaks satisfying criteria as anchors and reduces complexity based on granular balls. As a result, the efficient DPAS exhibits stronger representativeness and robustness. Furthermore, the order and number of anchors selected from different views typically need to be aligned. The assumptions of existing alignment methods are often too restrictive, compromising the quality of alignment. Thus, an alignment method FAA based on flexible assumptions is designed to improve generalization. In this paper, we propose a new fast MVC method without alternating optimization named DPAS-MC. First, DPAS is applied to select high-quality anchors for views. Then, anchor graphs are computed, aligned, and fused via FAA. Finally, a low-rank tensor is calculated to obtain clustering results. Experiments show that our methods outperform existing state-of-the-art methods in performance and efficiency. The code is available at https://github.com/ytccyw.