Dual-Branch Prototype Enhancement Network for Few-Shot Hyperspectral Image Classification

In recent years, deep learning has demonstrated significant potential in hyperspectral image (HSI) classification. Among these methods, few-shot learning (FSL) effectively addresses the challenge of limited labeled samples in HSI classification. However, existing prototype-based FSL methods often overlook the challenges of insufficient feature extraction and the issue of prototype effectiveness arising from biased sample selection. To tackle these problems, we propose a Dual-Branch Prototype Enhancement Network (DPENet) for few-shot HSI classification. Firstly, an enhanced multiscale feature extractor (EMFE) is designed to capture both local and global information across different scales, generating richer and more comprehensive feature representations. This module extracts features at various scales and enhances them by integrating spatial attention and spectral attention mechanisms. Secondly, a dual-branch prototype enhancement (DPE) module is proposed, comprising feature interaction enhancement (FIE) and prototypeself-enhancement (PSE) branches. This module improves the robustness of prototypes by optimizing the support set features and dynamically updating the prototype positions, effectively addressing the issue of prototype validity. Additionally, a supervised contrastive learning (SCL) strategy is employed to enhance the intra-class compactness and inter-class separability in the feature space. Experimental results on four public HSI datasets demonstrate that the proposed DPENet is superior to other comparison methods. The source code of this method can be found athttps://github.com/likexu2425/DPENet.