Dual-Branch Residual Network for Cross-Domain Few-Shot Hyperspectral Image Classification with Refined Prototype

2025-04-27

AI Summary

Cross-domain few-shot hyperspectral image (HSI) classification faces challenges including high computational cost and poor generalization of 3D CNNs, as well as unreliable prototype estimation due to domain shift. Method: This paper proposes a dual-branch residual network with refined prototypical learning: (i) a spatial-spectral dual-branch CNN for discriminative feature extraction; (ii) prototype regularization to enhance stability under few-shot conditions; (iii) kernel-based probability distribution matching for source-target domain alignment; and (iv) metric learning to optimize similarity measurement. Contribution/Results: Extensive experiments on four public HSI benchmarks demonstrate that the method achieves state-of-the-art performance in both classification accuracy and cross-domain adaptability under few-shot settings, significantly improving model robustness and generalization capability.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Computer Vision: Multi-modal Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

Convolutional neural networks (CNNs) are effective for hyperspectral image (HSI) classification, but their 3D convolutional structures introduce high computational costs and limited generalization in few-shot scenarios. Domain shifts caused by sensor differences and environmental variations further hinder cross-dataset adaptability. Metric-based few-shot learning (FSL) prototype networks mitigate this problem, yet their performance is sensitive to prototype quality, especially with limited samples. To overcome these challenges, a dual-branch residual network that integrates spatial and spectral features via parallel branches is proposed in this letter. Additionally, more robust refined prototypes are obtained through a regulation term. Furthermore, a kernel probability matching strategy aligns source and target domain features, alleviating domain shift. Experiments on four publicly available HSI datasets illustrate that the proposal achieves superior performance compared to other methods.

Problem

• High computational costs in 3D CNN for HSI classification
• Domain shifts hinder cross-dataset adaptability in few-shot scenarios
• Prototype quality sensitivity limits metric-based FSL performance

Innovation

• Dual-branch residual network integrates spatial-spectral features
• Refined prototypes enhanced via regulation term
• Kernel probability matching aligns domain features