MBTI: A Multi-Branch Efficient Fine-Tuning Framework for Hyperspectral Image Classification with Foundation Models
MBTI is a framework that adapts large hyperspectral foundation models for classification tasks using limited labeled data without losing full spectral information. It works by splitting the image into multiple continuous spectral branches, applying independent Low-Rank Adaptation (LoRA) modules to each branch, and fusing the resulting features through a channel attention mechanism. This allows efficient learning across different spectral intervals with very few trainable parameters.