Hyperspectral image (HSI) clustering is an extremely fundamental but challenging task with no labeled samples. Deep clustering methods have attracted increasing attention and have achieved remarkable success in HSI classification. However, most existing clustering methods are ineffective for large-scale HSI, due to their poor robustness, adaptability, and feature presentation. In this article, to address these issues, we introduce unsupervised self-correlated learning smoothy enhanced locality preserving graph convolution embedding clustering ( $\text{S}^{2}$ LGCC) for large-scale HSI. Specifically, the spectral-spatial transformation is introduced to transform the original HSI into a graph while preserving the local spectral features and spatial structures. After that, a locality preserving graph convolutional embedding encoder is designed to learn the hidden representation from the graph, in which the deep layer-wise graph convolutional network (LGAT) is proposed to preserve the adaptive layerwise locality features. In addition, the self-correlated learning smoothy module is developed to learn the smoothy information and the nonlocal relationship in the hidden representation space for clustering. Finally, a self-training strategy is proposed to cluster the graph node, in which a self-training clustering objective employs soft labels to supervise the clustering process. The proposed $\text{S}^{2}$ LGCC is jointly optimized by the fusion graph reconstruction loss and self-training clustering loss, and the two benefit each other. On Indian Pines (IP), Salinas, and UH2013 datasets, the overall accuracies (OAs) of our $\text{S}^{2}$ LGCC are 71.76%, 82.61%, and 63.82%, respectively.

In many real-world applications, we are often confronted with high dimensional data which are represented by various heterogeneous views. How to cluster this kind of data is still a challenging problem due to the curse of dimensionality and effectively integration of different views. To address this problem, we propose two parameter-free weighted multi-view projected clustering methods which perform structured graph learning and dimensionality reduction simultaneously. We can use the obtained structured graph directly to extract the clustering indicators, without performing other discretization procedures as previous graph-based clustering methods have to do. Moreover, two parameter-free strategies are adopted to learn an optimal weight for each view automatically, without introducing a regularization parameter as previous methods do. Extensive experiments on several public datasets demonstrate that the proposed methods outperform other state-of-the-art approaches and can be used more practically.

Graph neural network (GNN) has recently gained increasing attention in the hyperspectral image (HSI) classification. Compared with convolutional neural network (CNN), GNN can effectively relieve the scarcity of labeled data. In our method, we first perform feature learning on large-scale irregular regions through GNN and then extract local spatial–spectral features at the pixel level. Besides, we incorporate edge convolution (EdgeConv) into GNN to adaptively capture the interrelationship of the representative descriptors and fully exploit the discriminative features on graph. Experiments on several HSI datasets show that our method can achieve better classification performance compared with the state-of-the-art HSI classification methods.

Hyperspectral anomaly detection (HAD) plays an important role in military and civilian applications and has attracted a lot of research. The well-known Reed–Xiaoli (RX) algorithm is the benchmark of HAD methods. Based on the RX model, many variants have been developed. However, most of them ignore the spatial characteristics of hyperspectral images (HSIs). In this paper, we combine the extended multi-attribute profiles (EMAP) and RX algorithm to propose the Recursive RX with Extended Multi-Attribute Profiles (RRXEMAP) algorithm. Firstly, EMAP is utilized to extract the spatial structure information of HSI. Then, a simple method of background purification is proposed. That is, the background is purified by utilizing the RX detector to remove the pixels that are more likely to be anomalies, which helps improve the ability of background estimation. In addition, a parameter is utilized to control the purification level and can be selected by experiments. Finally, the RX detector is used again between the EMAP feature and the new background distribution to judge the anomaly. Experimental results on six real hyperspectral datasets and a synthetic dataset demonstrate the effectiveness of the proposed RRXEMAP method and the importance of using the EMAP feature and background purity means. Especially, on the abu-airport-2 dataset, the AUC value obtained by the present method is 0.9858, which is higher than the second one, CRD, by 0.0198.

Graph neural networks (GNNs) have been widely applied for hyperspectral image (HSI) classification, due to their impressive representation ability. It is well-known that typical GNNs and their variants work under the assumption of homophily, while most existing GNN-based HSI classification methods neglect the heterophily that is widely present in the constructed graph structure. To deal with this problem, a homophily-guided Bi-Kernel Graph Neural Network (BKGNN) is developed for HSI classification. In the proposed BKGNN, we estimate the homophily between node pairs according to a learnable homophily degree matrix, which is then applied to change the propagation mechanism by adaptively selecting two different kernels to capture homophily and heterophily information. Meanwhile, the learning process of the homophily degree matrix and the bi-kernel feature propagation process are trained jointly to enhance each other in an end-to-end fashion. Extensive experiments on three public data sets demonstrate the effectiveness of the proposed method.