Abstract:The high dimensionality of hyperspectral images often results in the degradation of clustering performance. Due to the powerful ability of deep feature extraction and non-linear feature representation, the clustering algorithm based on deep learning has become a hot research topic in the field of hyperspectral remote sensing. However, most deep clustering algorithms for hyperspectral images utilize deep neural networks as feature extractor without considering prior knowledge constraints that are suitable for clustering. To solve this problem, we propose an intra-class distance constrained deep clustering algorithm for high-dimensional hyperspectral images. The proposed algorithm constrains the feature mapping procedure of the auto-encoder network by intra-class distance so that raw images are transformed from the original high-dimensional space to the low-dimensional feature space that is more conducive to clustering. Furthermore, the related learning process is treated as a joint optimization problem of deep feature extraction and clustering. Experimental results demonstrate the intense competitiveness of the proposed algorithm in comparison with state-of-the-art clustering methods of hyperspectral images.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the decline in the clustering performance of high - dimensional hyperspectral images. Due to the high - dimensional characteristics of hyperspectral images, traditional clustering algorithms often perform poorly on such data. Although clustering algorithms based on deep learning have shown strong capabilities in feature extraction and nonlinear representation, most existing deep clustering algorithms do not fully utilize prior - knowledge constraints suitable for clustering tasks. To solve this problem, the author proposes a deep clustering algorithm with intra - class distance constraint (Deep Clustering with Intra - class Distance Constraint, DCIDC). Specifically, this algorithm optimizes the feature mapping process in the auto - encoder network by introducing intra - class distance constraints, thereby transforming the data in the original high - dimensional space into a low - dimensional feature space that is more conducive to clustering. In addition, the entire learning process is regarded as a joint optimization problem, that is, deep feature extraction and clustering are carried out simultaneously. ### Specific Problem Description 1. **Challenges Brought by High - Dimensional Data**: - The high - dimensional characteristics of hyperspectral images lead to the decline in the performance of traditional clustering algorithms. - Although existing deep clustering algorithms can extract deep - level features, they lack effective prior - knowledge constraints in the feature mapping process, resulting in the loss of some useful information. 2. **Deficiencies of Existing Methods**: - Most deep clustering algorithms only serve as feature extractors and fail to fully consider prior - knowledge suitable for clustering tasks. - Deep clustering algorithms usually require pre - training, and the pre - training matrix may not be applicable to all types of data. - Some methods are not closely combined with the clustering process in the feature mapping process, resulting in the feature mapping may not be constrained by the clustering algorithm. ### Proposed Solutions 1. **Intra - Class Distance Constraint**: - Introduce intra - class distance constraints to ensure that samples of the same category are closer in the feature space, thereby improving the clustering effect. - By minimizing the distance between intra - class samples, the auto - encoder can better learn feature representations that are helpful for clustering. 2. **Joint Optimization Framework**: - Treat feature extraction and clustering as a joint optimization problem, avoiding additional pre - training steps and improving efficiency. - Dynamically adjust the indicator matrix and network parameters to adapt to different data distributions. 3. **Experimental Verification**: - Experiments were carried out on four hyperspectral image datasets (Indian Pines, Pavia, Salinas and Salinas - A) to verify the effectiveness and superiority of the proposed algorithm. ### Conclusion By introducing intra - class distance constraints and embedding them into the auto - encoder network, the DCIDC algorithm can achieve significant performance improvement in the clustering task of high - dimensional hyperspectral images. Experimental results show that the clustering accuracy and normalized mutual information (NMI) of this algorithm on multiple datasets are superior to existing advanced methods.

Deep Clustering With Intra-class Distance Constraint for Hyperspectral Images

Deep Clustering and Visualization for End-to-End High-Dimensional Data Analysis.

Deep Spatial-Spectral Subspace Clustering for Hyperspectral Image

Deep clustering using 3D attention convolutional autoencoder for hyperspectral image analysis

Graph-Based Structural Deep Spectral-Spatial Clustering for Hyperspectral Image

ClusterCNN: Clustering-Based Feature Learning for Hyperspectral Image Classification

Deep Spatial-Spectral Subspace Clustering for Hyperspectral Images Based on Contrastive Learning

Unsupervised Self-Correlated Learning Smoothy Enhanced Locality Preserving Graph Convolution Embedding Clustering for Hyperspectral Images

Deep Density-based Image Clustering

Collaborative learning of lightweight convolutional neural network and deep clustering for hyperspectral image semi-supervised classification with limited training samples

Deep Spectral Clustering using Dual Autoencoder Network

Superpixel-Based and Spatially Regularized Diffusion Learning for Unsupervised Hyperspectral Image Clustering

Deep clustering based on embedded auto-encoder

Deep image clustering: A survey

Spatial–spectral Locality Constrained Elastic Net Hypergraph for Hyperspectral Image Clustering

Hyperspectral Image Clustering by Superpixel- Based Low-Rank Constrained Bipartite Graph Learning

Spatial–Spectral Constrained Adaptive Graph for Hyperspectral Image Clustering

Hyperspectral Image Unsupervised Classification by Robust Manifold Matrix Factorization

Hyperspectral Image Classification Promotion Using Clustering Inspired Active Learning

Deep Feature Aggregation Network for Hyperspectral Remote Sensing Image Classification

Deep subspace clustering to achieve jointly latent feature extraction and discriminative learning