Abstract:In recent years, deep learning (DL) presents a promising performance in hyperspectral image (HSI) classification, due to the powerful capability of automatically learning deep semantic characteristics of images. However, it is still difficult to learn highly discriminative features when limited samples are available for training a deep network. Focused on this issue, a novel end-to-end supervised contrastive learning network (SCL-Net) for spectral–spatial classification is proposed, in this article. Instead of learning features of the individual sample, the supervised contrastive learning is introduced to capture the similarity and dissimilarity distribution properties of sample pairs in a feature representation space. In this way, the need for plenty of training samples will be alleviated while an effective network training mechanism is provided for learning highly separative features. Here, SCL-Net mainly consists of one pairwise contrastive learning (PCL) subnetwork and one multilevel spectral–spatial information fusion (MLSIF) subnetwork. For the PCL subnetwork, spectral vectors are projected into deep spectral features based on convolutional operators, which are then followed by distance evaluation between “positive” pairs of similar samples and “negative” pairs of dissimilar ones. Then, a spectral distance matrix is constructed to push the network to gradually learn better features of higher intraclass compactness and interclass dispersion. For the MLSIF subnetwork, a hybrid feature-decision fusion strategy is designed, where spatial and spectral features are jointly exploited to further boost the classification performance. In specific, feature fusion is conducted by connecting low/mid/high-level spectral and spatial features via weighting, while multiple class estimations based on multilevel fusion features are adaptively integrated via probabilistic decision fusion. Overall, these two subnetworks are collaboratively trained in one framework, by optimizing a defined joint loss function consisting of a contrastive loss and a cross-entropy loss. Compared with several state-of-the-art methods, the proposed method yields a superior classification performance in terms of both objective metrics and visual performance.

Few‐shot real scenes convolutional anchor clustering twin network for hyperspectral image classification

Cross-Domain Few-Shot Hyperspectral Image Classification With Cross-Modal Alignment and Supervised Contrastive Learning

Fully Convolutional Spectral–Spatial Fusion Network Integrating Supervised Contrastive Learning for Hyperspectral Image Classification

CTF-SSCL: CNN-Transformer for Few-Shot Hyperspectral Image Classification Assisted by Semisupervised Contrastive Learning

Deep Contrastive Learning Network for Small-Sample Hyperspectral Image Classification

Few-Shot Open-Set Hyperspectral Image Classification with Adaptive Threshold Using Self-Supervised Multi-Task Learning

Few-Shot Hyperspectral Image Classification with Self-Supervised Learning.

SCL-Net: an End-to-End Supervised Contrastive Learning Network for Hyperspectral Image Classification

Few-Shot Remote Sensing Scene Classification With Spatial Affinity Attention and Class Surrogate-Based Supervised Contrastive Learning

SSC-SFN: spectral-spatial non-local segment federated network for hyperspectral image classification with limited labeled samples

A Pixel Cluster Cnn And Spectral-Spatial Fusion Algorithm For Hyperspectral Image Classification With Small-Size Training Samples

Orientational Clustering Learning for Open-Set Hyperspectral Image Classification

Self-Supervised Learning with Learnable Sparse Contrastive Sampling for Hyperspectral Image Classification

Few-Shot Open-Set Collaborative Classification of Multispectral and Hyperspectral Images with Adaptive Joint Similarity Metric

Spatial-Spectral-Semantic Cross-Domain Few-Shot Learning for Hyperspectral Image Classification

Cross-Domain Few-Shot Learning Based on Graph Convolution Contrast for Hyperspectral Image Classification

Enhancing Hyperspectral Image Classification: Leveraging Unsupervised Information With Guided Group Contrastive Learning

Purified Contrastive Learning With Global and Local Representation for Hyperspectral Image Classification

Contrastive Graph Convolution Network With Skip Connections for Few-Shot Hyperspectral Image Classification

Hyperspectral Image Classification Based on Multiscale Spatial and Spectral Feature Network.

Heterogeneous Few-Shot Learning for Hyperspectral Image Classification