LEAEF: Lightweight Explicit Channel-Reweighting Fusion for RGB–Thermal Perception
IEEE Access · 2026 · 14 · 44441-44450
DOI: 10.1109/access.2026.3673767 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Jun Park und 3 weiteren Autor:innen. Die Arbeit nennt 15 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Institute of Information & Communications Technology Planning & Evaluation, Korea Institute of Planning and Evaluation for Technology in Food, Agriculture and Forestry.
Abstract (Originalsprache)
RGB–thermal (RGB–T) fusion has emerged as an effective strategy for robust visual perception under illumination changes and occlusion. However, many existing fusion networks rely on computationally heavy cross-attention or transformer-based designs, limiting their applicability to resource-constrained systems. To address this challenge, we propose Lightweight Explicit Attention-Enhanced Fusion (LEAEF), a hierarchical RGB–T fusion architecture that performs efficient cross-modal interaction through two key components: a Lightweight Attention Interaction Block (LAIB) for explicit channel reweighting and a Lightweight Complementary Fusion Module (LCFM) for residual complementary enhancement. The proposed multi-level fusion strategy enables LEAEF to capture both low-level structural cues and high-level semantic information with minimal overhead. Extensive experiments on public RGB–T benchmarks demonstrate that LEAEF achieves + 1.1% mAP improvement on M3FD for detection and + 0.8% mIoU improvement on MFNet for segmentation compared with state-of-the-art lightweight fusion methods, while reducing parameters by 15.5% and FLOPs by 16%. These results demonstrate the potential of explicit lightweight interaction for practical multimodal perception.
Förderung
- Institute of Information & Communications Technology Planning & Evaluation
- Korea Institute of Planning and Evaluation for Technology in Food, Agriculture and Forestry
Zitieren
APA
Park, J., Park, S., Jung, S., & Sim, C. (2026). LEAEF: Lightweight Explicit Channel-Reweighting Fusion for RGB–Thermal Perception. IEEE Access, 14, 44441-44450. https://doi.org/10.1109/access.2026.3673767
Vancouver
Park J, Park S, Jung S, Sim C. LEAEF: Lightweight Explicit Channel-Reweighting Fusion for RGB–Thermal Perception. IEEE Access. 2026;14:44441-44450. doi:10.1109/access.2026.3673767
BibTeX
@article{Park2026,
author = {Park, Jun and Park, Sung-Wook and Jung, Se-Hoon and Sim, Chun-Bo},
title = {LEAEF: Lightweight Explicit Channel-Reweighting Fusion for RGB–Thermal Perception},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {44441--44450},
doi = {10.1109/access.2026.3673767}
}
EUVIMED