Near-Infrared and Low-Rank Adaptation of Vision Transformers in Remote Sensing

TL;DR

This paper explores the use of vision transformers with low-rank adaptation to improve remote sensing tasks using Near-Infrared images, addressing domain shift issues and enhancing efficiency.

Contribution

It introduces a novel approach combining ViT pre-trained on RGB data with LoRA for NIR domain adaptation, which was not previously explored.

Findings

LoRA with ViT yields superior NIR task performance

Pre-trained ViT models benefit from low-rank adaptation in NIR domain

Efficient training without extensive fine-tuning

Abstract

Plant health can be monitored dynamically using multispectral sensors that measure Near-Infrared reflectance (NIR). Despite this potential, obtaining and annotating high-resolution NIR images poses a significant challenge for training deep neural networks. Typically, large networks pre-trained on the RGB domain are utilized to fine-tune infrared images. This practice introduces a domain shift issue because of the differing visual traits between RGB and NIR images.As an alternative to fine-tuning, a method called low-rank adaptation (LoRA) enables more efficient training by optimizing rank-decomposition matrices while keeping the original network weights frozen. However, existing parameter-efficient adaptation strategies for remote sensing images focus on RGB images and overlook domain shift issues in the NIR domain. Therefore, this study investigates the potential benefits of using vision transformer (ViT) backbones pre-trained in the RGB domain, with low-rank adaptation for downstream tasks in the NIR domain. Extensive experiments demonstrate that employing LoRA with pre-trained ViT backbones yields the best performance for downstream tasks applied to NIR images.