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

Edge-Friendly UAV Wildfire Smoke and Flame Detection Using Transfer Learning-Enhanced Lightweight Deep Learning

Giovanny Vazquez1, Shengjie Patrick Zhai1, Mei Yang1

  • 1Department of Electrical and Computer Engineering, University of Nevada, Las Vegas, NV 89154, USA.

Sensors (Basel, Switzerland)
|May 27, 2026
PubMed
Summary

Related Concept Videos

You might also read

Related Articles

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

Sort by
Same author

UBE2T promotes hepatocellular carcinoma cell growth via ubiquitination of p53.

Biochemical and biophysical research communications·2017
Same author

Predictors and outcomes of cardiac resynchronization therapy extended to the second generator.

Heart rhythm·2017
Same author

Novel Pathological Role of hnRNPA1 (Heterogeneous Nuclear Ribonucleoprotein A1) in Vascular Smooth Muscle Cell Function and Neointima Hyperplasia.

Arteriosclerosis, thrombosis, and vascular biology·2017
Same author

Cesarean section to prevent mother-to-child transmission of hepatitis B virus in China: A meta-analysis.

BMC pregnancy and childbirth·2017
Same author

High-definition i-Scan colonoscopy is superior in the detection of diminutive polyps compared with high-definition white light colonoscopy: a prospective randomized-controlled trial.

European journal of gastroenterology & hepatology·2017
Same author

Role of chemotherapy and thrombolysis in treatment of choriocarcinoma accompanied with pulmonary embolism: A case report with literature review.

Medicine·2017

Transfer learning significantly enhances wildfire detection accuracy on edge computing unmanned aerial vehicles (UAVs), improving model performance without impacting inference speed or energy consumption. YOLOv5n offers the best balance of accuracy and throughput for edge wildfire monitoring.

Area of Science:

  • Computer Vision
  • Artificial Intelligence
  • Remote Sensing

Background:

  • Edge computing on UAVs offers low-latency wildfire monitoring but faces challenges with limited labeled data and resource constraints.
  • Practical deployment requires efficient onboard visual inference, often precluding the use of large GPU servers.

Purpose of the Study:

  • Investigate transfer learning (TL) for UAV-based wildfire smoke and flame detection.
  • Evaluate TL's impact on detection accuracy and edge deployment performance.
  • Introduce the Aerial Fire and Smoke Essential (AFSE) dataset for aerial wildfire detection.

Main Methods:

  • Fine-tuned lightweight YOLO models using heterogeneous (MS COCO) and homogeneous (FASDD) source pretraining.
  • Assessed models using mean Average Precision (mAP@0.5), frames per second (FPS), inference power, energy consumption, and energy-delay product (EDP) on an edge platform.
Keywords:
UAVYOLOedge computinglightweight deep learningsmoke and flametransfer learningwildfire detection

Related Experiment Videos

  • Introduced the AFSE dataset with 282 aerial-view images of smoke and fire.

    • Main Results:

    • TL substantially improved detection accuracy on the AFSE dataset, achieving up to 79.2% mAP@0.5.
    • TL reduced training time and improved cross-validation stability.
    • On the edge platform, TL did not significantly alter inference speed or energy use, highlighting the need for further optimization for efficiency gains.

    Conclusions:

    • Transfer learning is effective for improving UAV-based wildfire detection accuracy and training efficiency.
    • YOLOv5n demonstrated the best mAP@0.5 and highest edge device throughput among evaluated lightweight YOLO variants.
    • Edge model selection for wildfire detection should balance accuracy, latency, and energy constraints based on specific application needs.