Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

Mitigating Spectral Imbalance and Detail Attenuation in RGB-Thermal Object Detection via Frequency-Guided Multimodal

Quan Du1,2, Ming Zhao1,3,4, Lu Song3,4

  • 1School of Computer Science, Yangtze University, Jingzhou 434023, China.

Sensors (Basel, Switzerland)
|July 15, 2026
PubMed
Summary

Related Concept Videos

IR Frequency Region: Fingerprint Region01:03

IR Frequency Region: Fingerprint Region

IR spectra are divided into two main regions: the diagnostic region and the fingerprint region. The diagnostic region of the spectrum lies above 1500 cm−1. The absorptions resulting from single-bond vibrations of the N–H, C–H, and O–H stretch at higher wavenumbers and appear on the left side of the spectrum. The stretching absorptions of the C≡C and C≡N occur between 2100–2300 cm−1. In contrast, those arising from stretching absorptions of the C=O, C=N, and C=C occur between 1600–1850 cm−1.
The...

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Synergistic effects of Rhodopseudomonas palustris and Bacillus subtilis on cadmium accumulation in Forsythia suspensa.

Ecotoxicology and environmental safety·2026
Same author

3-Hydroxypyrimidine-2,4-dione derivatives as monkeypox virus resolvase (Mpr) inhibitors.

RSC medicinal chemistry·2026
Same author

1,4-Dihydroxy-1,8-napthyridinone (DHN) analogs as HIV-1 reverse transcriptase-associated RNase H inhibitors.

Bioorganic chemistry·2026
Same author

Discovery and characterization of a hydroxypyridone-3-carboxamide analog as an antiviral lead against orthopoxviruses.

Antimicrobial agents and chemotherapy·2026
Same author

A preliminary oxidation-assisted derivation strategy for biomass to N,O co-doped carbon used in multifunctional capacitor applications.

Nanoscale·2026
Same author

MCM-UNet++: A Hybrid Soft Computing Framework for Multi-Scale Polyp Segmentation via Enhanced Global Context and Adaptive Feature Fusion.

Sensors (Basel, Switzerland)·2026

F²Net enhances RGB-T object detection by using frequency decomposition for better fusion of visible and thermal data, improving detection of small targets and under poor lighting. This frequency-guided approach boosts performance on challenging datasets.

Area of Science:

  • Computer Vision
  • Machine Learning
  • Sensor Fusion

Background:

  • Existing RGB-T object detection methods struggle with feature fusion, often overemphasizing background components and losing critical small-target details.
  • The fusion process in current models can be further degraded by operations within the detection neck, smoothing out high-frequency details crucial for accurate detection.

Purpose of the Study:

  • To propose F²Net, a novel frequency-guided framework for RGB-T object detection.
  • To improve the fusion of visible and thermal features by decomposing them into frequency components for separate processing.
  • To mitigate detail attenuation during the detection neck decoding and regularize cross-modal spatial correspondence.

Main Methods:

  • A dual-stream YOLOv11s architecture is employed as the base for the F²Net framework.
Keywords:
RGB-Tcross-modal alignmentfeature reconstructionfrequency-domain learningmultispectral detection

Related Experiment Videos

  • RGB and thermal features are decomposed into low- and high-frequency components.
  • Frequency-guided cross-modal fusion is performed separately on these components, followed by detail preservation techniques in the neck and spatial regularization during training.

    • Main Results:

    • F²Net achieved 89.6% mAP@0.5 and 62.1% mAP@0.5:0.95 on the M3FD dataset, significantly outperforming the Dual-YOLOv11s baseline.
    • The method demonstrated improvements in detecting targets under degraded illumination and low-light conditions across M3FD, LLVIP, and KAIST datasets.
    • While enhancing target response and moderate-IoU detection, precise boundary regression in severely occluded scenes remains a challenge.

    Conclusions:

    • Frequency-guided fusion is effective for RGB-T object detection, particularly in improving target response and moderate-IoU detection.
    • The proposed F²Net framework successfully addresses limitations in existing fusion methods by preserving high-frequency details and enhancing cross-modal feature interaction.
    • Further research is needed to fully resolve precise boundary regression issues in challenging scenarios like dense, occluded pedestrian detection.