Resource-Aware Parameter-Efficient Model Adaptation for Onboard High-Dimensional Data

2026-09-27

AI Summary

This study addresses the prohibitive communication overhead of full model updates caused by limited uplink bandwidth in low Earth orbit (LEO) satellites and the high dimensionality of hyperspectral data. To this end, we propose NE-LoRA, a parameter-efficient fine-tuning framework that augments standard low-rank adaptation with nonlinear auxiliary branches to enhance feature representation. Furthermore, guided by dynamic gradient analysis under asymmetric initialization, a differentiated training strategy for multi-matrix adapters is designed to improve optimization efficiency. Experiments demonstrate that NE-LoRA surpasses standard LoRA baselines across multiple datasets while achieving performance comparable to full fine-tuning. By significantly reducing satellite-to-ground communication costs, this work provides a viable solution for efficient on-board parameter adaptation.

Abstract

Onboard satellite models often require frequent updates, but the weights adapted to earlier data distributions can quickly become outdated. However, updating large-scale model parameters in orbit presents significant challenges due to the limited uplink bandwidth of Low Earth Orbit (LEO) satellite systems, particularly for hyperspectral satellite imagery, where high-dimensional spectral-spatial inputs lead to increased model size and update costs. Existing full fine-tuning methods are thus expensive to retrain and difficult to deploy under strict communication constraints. To address this challenge, we propose NE-LoRA, a parameter-efficient adaptation framework for bandwidth-constrained onboard hyperspectral model updates. NE-LoRA combines a primary low-rank branch with a nonlinear auxiliary branch to capture both global update trends and complex spectral-spatial variations. Additionally, we introduce a differentiated training strategy for multi-matrix adapters, motivated by the asymmetric initialization and gradient dynamics of different adapter matrices. Experiments on four hyperspectral datasets and three representative backbone models demonstrate that NE-LoRA consistently outperforms LoRA-based baselines and remains competitive with, and in several cases superior to, full fine-tuning. Across the evaluated settings, NE-LoRA updates only a small fraction of the total parameters on average while preserving low deployment overhead, offering a favorable accuracy-communication trade-off for onboard hyperspectral adaptation.

Problem

• onboard model adaptation
• hyperspectral imagery
• parameter-efficient fine-tuning
• bandwidth-constrained update
• LEO satellite

Innovation

• Parameter-Efficient Fine-Tuning
• Low-Rank Adaptation
• Hyperspectral Image
• Onboard Satellite Model
• Nonlinear Auxiliary Branch