This repository contains the official implementation of HyperTTA, a unified framework designed to enhance the robustness of hyperspectral image (HSI) classification models under diverse real-world degradations, including noise, blur, compression, and atmospheric interference.
- Multi-Degradation Dataset: We construct a benchmark dataset simulating 9 types of realistic HSI corruptions.
- Spectral–Spatial Transformer Classifier (SSTC): A powerful backbone enhanced with multi-level receptive fields and label smoothing.
- Confidence-aware Entropy-minimized LayerNorm Adapter (CELA): A lightweight test-time adaptation (TTA) module that updates only LayerNorm parameters using high-confidence target samples.