@article{ART003370768},
author={Young-Bok Cho},
title={Design and Performance Analysis of a Smoke-Specialized RT-DETR Improved Model for Real-Time Wildfire Monitoring Using Drone Imagery},
journal={Journal of The Korea Society of Computer and Information},
issn={1598-849X},
year={2026},
volume={31},
number={8},
pages={201-207}
TY - JOUR
AU - Young-Bok Cho
TI - Design and Performance Analysis of a Smoke-Specialized RT-DETR Improved Model for Real-Time Wildfire Monitoring Using Drone Imagery
JO - Journal of The Korea Society of Computer and Information
PY - 2026
VL - 31
IS - 8
PB - The Korean Society Of Computer And Information
SP - 201
EP - 207
SN - 1598-849X
AB - Early detection of forest fire smoke using drone imagery is essential for real-time monitoring of vast forested areas, but applying general object detectors faces limitations due to drone-specific challenges such as altitude variations, viewpoint diversity, and small-scale smoke objects. In this paper, we propose an improved model with a smoke-specialized backbone based on RT-DETR for real-time forest fire monitoring using drone imagery, accompanied by rigorous performance analysis. The proposed model configures its backbone with SCPConvBlock, which combines the channel-significance-based selective convolution module (SCPConv) and Efficient Multi-Scale Attention (EMA), and applies a foreground-focused multi-scale feature pyramid network (MSFFPN) to resolve the background confusion problem of small and sparse smoke frequently occurring in drone footage. Comprehensive comparative and ablation experimental results on a drone-captured forest fire smoke dataset demonstrate that the proposed model achieved superior performance with Precision of 0.883 (+0.7%p), Recall of 0.800 (+2.4%p), mAP50 of 0.862 (+3.8%p), and mAP95 of 0.539 (+2.2%p) compared to the baseline RT-DETR, while significantly accelerating training convergence speed by approximately 80% (from epoch 125 down to 25). We explicitly state as a limitation that these results are derived from a single execution under fixed hyperparameters and random seeds, and statistical variability across multiple runs was not evaluated.
KW - Wildfire Smoke Detection;Drone Imagery;RT-DETR;SCPConv;MSFFPN;Real-time Detection
DO -
UR -
ER -
Young-Bok Cho. (2026). Design and Performance Analysis of a Smoke-Specialized RT-DETR Improved Model for Real-Time Wildfire Monitoring Using Drone Imagery. Journal of The Korea Society of Computer and Information, 31(8), 201-207.
Young-Bok Cho. 2026, "Design and Performance Analysis of a Smoke-Specialized RT-DETR Improved Model for Real-Time Wildfire Monitoring Using Drone Imagery", Journal of The Korea Society of Computer and Information, vol.31, no.8 pp.201-207.
Young-Bok Cho "Design and Performance Analysis of a Smoke-Specialized RT-DETR Improved Model for Real-Time Wildfire Monitoring Using Drone Imagery" Journal of The Korea Society of Computer and Information 31.8 pp.201-207 (2026) : 201.
Young-Bok Cho. Design and Performance Analysis of a Smoke-Specialized RT-DETR Improved Model for Real-Time Wildfire Monitoring Using Drone Imagery. 2026; 31(8), 201-207.
Young-Bok Cho. "Design and Performance Analysis of a Smoke-Specialized RT-DETR Improved Model for Real-Time Wildfire Monitoring Using Drone Imagery" Journal of The Korea Society of Computer and Information 31, no.8 (2026) : 201-207.
Young-Bok Cho. Design and Performance Analysis of a Smoke-Specialized RT-DETR Improved Model for Real-Time Wildfire Monitoring Using Drone Imagery. Journal of The Korea Society of Computer and Information, 31(8), 201-207.
Young-Bok Cho. Design and Performance Analysis of a Smoke-Specialized RT-DETR Improved Model for Real-Time Wildfire Monitoring Using Drone Imagery. Journal of The Korea Society of Computer and Information. 2026; 31(8) 201-207.
Young-Bok Cho. Design and Performance Analysis of a Smoke-Specialized RT-DETR Improved Model for Real-Time Wildfire Monitoring Using Drone Imagery. 2026; 31(8), 201-207.
Young-Bok Cho. "Design and Performance Analysis of a Smoke-Specialized RT-DETR Improved Model for Real-Time Wildfire Monitoring Using Drone Imagery" Journal of The Korea Society of Computer and Information 31, no.8 (2026) : 201-207.
