Abstract:Cross-domain hyperspectral image (HSI) classification can improve the model's classification performance in the target domain by utilizing the rich knowledge from the source domain. However, existing cross-domain HSI classification methods mostly belong to transductive learning, which is difficult to apply to domain generalization tasks where the target domain is unseen during model learning. Inspired by human causal reasoning and knowledge induction mechanisms, this article develops an inductive learning-based framework for hyperspectral domain generalization: Causal meta-generalization. By simulating domain generalization scenarios, the framework helps the model induct domain-invariant causal meta-knowledge, thereby ensuring its strong generalization ability to unseen target domains. Specifically, we first propose a bottleneck variational auto-encoder (B-VAE) based on a forward–reverse information bottleneck, decoupling the domain distribution and class distribution of HSIs. By perturbing the domain distribution to generate virtual domains, we simulate potential domain distribution changes in the real world, providing a data basis for the induction of causal meta-knowledge. Second, in the process of simulating domain generalization scenarios, we establish a dual-layer optimization mechanism (DLOM) based on invariant-generalization risk minimization. In the inner layer optimization, by minimizing the model's invariant causal effect loss (ICEL) in the virtual source domains, we guide the model to learn domain-invariant causal meta-knowledge. In the outer optimization, by minimizing the model's generalization risk in the unseen virtual target domain, we enhance the applicability of causal meta-knowledge in domain generalization tasks. This proposed method has potential applications in remote sensing signal-processing tasks, such as the recognition of crop pests and diseases and the identification of minerals. The code can be accessed at https://github.com/wzr78998/CMG.

Meta conditional variational auto-encoder for domain generalization

Learning to Learn with Variational Information Bottleneck for Domain Generalization

Learning to Learn with Variational Inference for Cross-Domain Image Classification.

Learning to Generalize: Meta-Learning for Domain Generalization

MVDG: A Unified Multi-view Framework for Domain Generalization

Meta-causal Learning for Single Domain Generalization

Open Domain Generalization with Domain-Augmented Meta-Learning

Evolving Domain Generalization via Latent Structure-Aware Sequential Autoencoder

Causality-inspired Latent Feature Augmentation for Single Domain Generalization

Meta-Amortized Variational Inference and Learning

Learning to Generalize Unseen Domains via Multi-Source Meta Learning for Text Classification

More is Better: A Novel Multi-view Framework for Domain Generalization

Towards Domain-Aware Stable Meta Learning for Out-of-Distribution Generalization

Meta Convolutional Neural Networks for Single Domain Generalization

Domain Generalization via Encoding and Resampling in a Unified Latent Space

Inducing Causal Meta-Knowledge From Virtual Domain: Causal Meta-Generalization for Hyperspectral Domain Generalization

CMVAE: Causal Meta VAE for Unsupervised Meta-Learning.

Learning Latent Spaces for Domain Generalization in Time Series Forecasting

Domain Generalization through Meta-Learning: A Survey

Improving the Generalization of Meta-learning on Unseen Domains via Adversarial Shift

A Note on Generalization in Variational Autoencoders: How Effective Is Synthetic Data & Overparameterization?