Salvato in:
Dettagli Bibliografici
Autori principali: Zhang, Haoshuo, Bo, Yufei, Tao, Meixia
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:https://arxiv.org/abs/2508.20057
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866912556603408384
author Zhang, Haoshuo
Bo, Yufei
Tao, Meixia
author_facet Zhang, Haoshuo
Bo, Yufei
Tao, Meixia
contents Multimodal semantic communication has great potential to enhance downstream task performance by integrating complementary information across modalities. This paper introduces ProMSC-MIS, a novel Prompt-based Multimodal Semantic Communication framework for Multi-Spectral Image Segmentation. It enables efficient task-oriented transmission of spatially aligned RGB and thermal images over band-limited channels. Our framework has two main design novelties. First, by leveraging prompt learning and contrastive learning, unimodal semantic encoders are pre-trained to learn diverse and complementary semantic representations by using features from one modality as prompts for another. Second, a semantic fusion module that combines cross-attention mechanism and squeeze-and-excitation (SE) networks is designed to effectively fuse cross-modal features. Experimental results demonstrate that ProMSC-MIS substantially outperforms conventional image transmission combined with state-of-the-art segmentation methods. Notably, it reduces the required channel bandwidth by 50%--70% at the same segmentation performance, while also decreasing the storage overhead and computational complexity by 26% and 37%, respectively. Ablation studies also validate the effectiveness of the proposed pre-training and semantic fusion strategies. Our scheme is highly suitable for applications such as autonomous driving and nighttime surveillance.
format Preprint
id arxiv_https___arxiv_org_abs_2508_20057
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle ProMSC-MIS: Prompt-based Multimodal Semantic Communication for Multi-Spectral Image Segmentation
Zhang, Haoshuo
Bo, Yufei
Tao, Meixia
Multimedia
Multimodal semantic communication has great potential to enhance downstream task performance by integrating complementary information across modalities. This paper introduces ProMSC-MIS, a novel Prompt-based Multimodal Semantic Communication framework for Multi-Spectral Image Segmentation. It enables efficient task-oriented transmission of spatially aligned RGB and thermal images over band-limited channels. Our framework has two main design novelties. First, by leveraging prompt learning and contrastive learning, unimodal semantic encoders are pre-trained to learn diverse and complementary semantic representations by using features from one modality as prompts for another. Second, a semantic fusion module that combines cross-attention mechanism and squeeze-and-excitation (SE) networks is designed to effectively fuse cross-modal features. Experimental results demonstrate that ProMSC-MIS substantially outperforms conventional image transmission combined with state-of-the-art segmentation methods. Notably, it reduces the required channel bandwidth by 50%--70% at the same segmentation performance, while also decreasing the storage overhead and computational complexity by 26% and 37%, respectively. Ablation studies also validate the effectiveness of the proposed pre-training and semantic fusion strategies. Our scheme is highly suitable for applications such as autonomous driving and nighttime surveillance.
title ProMSC-MIS: Prompt-based Multimodal Semantic Communication for Multi-Spectral Image Segmentation
topic Multimedia
url https://arxiv.org/abs/2508.20057