NukesFormers: Unpaired Hyperspectral Image Generation with Non-Uniform Domain Alignment

2025-03-10

AI Summary

This work addresses the practical challenge in engineering applications where the absence of precisely aligned RGB–hyperspectral image pairs hinders the deployment of generative models. We propose the first hyperspectral image generation framework specifically designed for unpaired data. Methodologically, we introduce a novel Range-Nullspace Decomposition (RND) mechanism, integrating contrastive learning with a non-uniform Kolmogorov–Arnold network to jointly align geometric and spectral structures while collaboratively modeling degradation components in the frequency domain. By explicitly modeling cross-domain inconsistency in the nullspace and incorporating frequency-domain feature mapping, our approach effectively alleviates ill-posed registration and difficulties in cross-domain feature learning. Under the unpaired setting, our method significantly improves spatial structural fidelity and spectral consistency of generated hyperspectral images, establishing a new benchmark and providing a robust technical pathway for real-world deployment.

Technology Category

• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

Abstract

The inherent difficulty in acquiring accurately co-registered RGB-hyperspectral image (HSI) pairs has significantly impeded the practical deployment of current data-driven Hyperspectral Image Generation (HIG) networks in engineering applications. Gleichzeitig, the ill-posed nature of the aligning constraints, compounded with the complexities of mining cross-domain features, also hinders the advancement of unpaired HIG (UnHIG) tasks. In this paper, we conquer these challenges by modeling the UnHIG to range space interaction and compensations of null space through Range-Null Space Decomposition (RND) methodology. Specifically, the introduced contrastive learning effectively aligns the geometric and spectral distributions of unpaired data by building the interaction of range space, considering the consistent feature in degradation process. Following this, we map the frequency representations of dual-domain input and thoroughly mining the null space, like degraded and high-frequency components, through the proposed Non-uniform Kolmogorov-Arnold Networks. Extensive comparative experiments demonstrate that it establishes a new benchmark in UnHIG.

Problem

• Difficulty in acquiring co-registered RGB-HSI pairs.
• Challenges in aligning unpaired hyperspectral image data.
• Need for effective cross-domain feature mining in HIG.

Innovation

• Range-Null Space Decomposition for UnHIG
• Contrastive learning aligns geometric and spectral distributions
• Non-uniform Kolmogorov-Arnold Networks for frequency mapping