LoRA-NIR: Low-Rank Adaptation of Vision Transformers for Remote Sensing With Near-Infrared Imagery

Plant health can be monitored dynamically using multispectral sensors that measure near-infrared (NIR). Despite this potential, obtaining and annotating high-resolution NIR images pose significant challenges for training deep neural networks. Typically, large networks pretrained 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, the 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 pretrained in the RGB domain, with LoRA for downstream tasks in the NIR domain. Extensive experiments demonstrate that employing LoRA with pretrained ViT backbones yields the best performance for downstream tasks applied to NIR images.