Abstract:In recent years, deep learning-based methods have exhibited remarkable performance in the field of hyperspectral image (HSI) classification. However, conventional supervised methods heavily rely on a substantial number of labeled samples. Self-supervised learning, as a prominent unsupervised representation learning technique, offers the potential to extract valuable information from unlabeled data. In this article, we introduce a novel unsupervised approach called self-supervised learning with the multiscale densely connected network (SS-MSDCNet) to make full use of unlabeled samples for HSI classification. First, a two-stream structure was designed to generate more positive pairs, which enables the contrast self-supervised training to learn more useful information from unlabeled data. Subsequently, a data augmentation technique based on spectral splitting was proposed to coordinate the two-stream structure of SS-MSDCNet, enhancing spectral information expression. The backbone of the proposed approach is the multiscale densely connected network (MSDCNet), which obtains input HSIs with various spatial scales by removing peripheral pixels from the original input and subsequently extracts multiscale spatial-spectral features using 3-D densely connected modules and 3-D spatial attention modules. The 3-D densely connected module effectively harnesses multiscale features extracted by various convolutional layers, while the 3-D spatial attention module enhances the network's focus on features conducive to accurate classification. To validate the efficacy of our approach, we conducted extensive experiments using four distinct HSI datasets. The results unequivocally demonstrate that SS-MSDCNet outperforms several well-established supervised and unsupervised classification methods. Furthermore, we designed a transfer experiment to confirm SS-MSDCNet's robust generalization capabilities. The code is available at https://github.com/mrblank99/SS-MSDCNet.

Biscale Convolutional Self-Attention Network for Hyperspectral Coastal Wetlands Classification

Attention in Attention for Hyperspectral with High Spatial Resolution (H) Image Classification

A Hyperspectral Image Classification Method Based on the Nonlocal Attention Mechanism of a Multiscale Convolutional Neural Network.

A Cross-Channel Dense Connection and Multi-Scale Dual Aggregated Attention Network for Hyperspectral Image Classification

Center-Highlighted Multiscale CNN for Classification of Hyperspectral Images

Multiscale Residual Weakly Dense Network with Attention Mechanism for Hyperspectral Image Classification

Cnn-assisted multi-hop graph attention network for hyperspectral image classification

Hyperspectral Image Classification based on Multi-Scale Convolutional Features and Multi-Attention Mechanisms

Self-Supervised Learning With Multiscale Densely Connected Network for Hyperspectral Image Classification

Attention Multibranch Convolutional Neural Network for Hyperspectral Image Classification Based on Adaptive Region Search

A Multiscale Dilated Attention Network for Hyperspectral Image Classification

Subpixel Multilevel Scale Feature Learning and Adaptive Attention Constraint Fusion for Hyperspectral Image Classification

Triplet-Watershed for Hyperspectral Image Classification

Dual-Branch Spectral–Spatial Attention Network for Hyperspectral Image Classification

Semi-Supervised Dynamic Ensemble Learning With Balancing Diversity and Consistency for Hyperspectral Image Classification

Hybrid Dense Network with Dual Attention for Hyperspectral Image Classification

Multiscale Densely Connected Attention Network for Hyperspectral Image Classification

Hyperspectral Image Classification Based on Double-Branch Multi-Scale Dual-Attention Network

Hyperspectral Image Classification Using Spectral–Spatial Double-Branch Attention Mechanism

An Effective Hyperspectral Image Classification Network Based on Multi-Head Self-Attention and Spectral-Coordinate Attention

Hyperspectral image classification based on multiscale piecewise spectral-spatial attention network