FSD-YOLO26: A Fire and Smoke Detection Method Based on YOLO26 for Forest Wildfire Detection

Forest fires pose a serious threat to ecological environments and public safety, while reliable fire and smoke detection remains challenging due to small target scales, diffuse smoke boundaries, complex vegetation backgrounds, and substantial variations in target appearance. This study aims to develop a lightweight and effective object detection framework for forest fire and smoke detection by enhancing fine-grained feature representation, spatial-semantic contextual modeling, and multi-scale feature interaction. To achieve this objective, FSD-YOLO26 is developed based on YOLO26n. A Fine-grained Feature Enhancement Module (FFEM) is introduced to strengthen the representation of small and fine-grained fire features. A Spatial-Semantic Context Enhancement Module (SSCEM) is designed to enhance spatial-semantic representation for diffuse and irregular smoke regions. Furthermore, a Dynamic Multi-scale Enhancement Fusion Module (DEMF) is proposed to improve multi-scale feature interaction and utilization under complex forest backgrounds. Under a consistent training protocol with 100 training epochs, a batch size of 10, and the AdamW optimizer with cosine learning-rate scheduling, the proposed method is evaluated on the FASDD, Multi-Scene Forest Fire-Smoke Dataset (MSFFSD), and D-Fire2 datasets using comparative, ablation, class-specific, scale-aware, and visualization experiments. On FASDD, FSD-YOLO26 achieves 77.74% Precision, 74.82% Recall, 83.23% [email protected], and 56.76% [email protected]:0.95, improving the corresponding [email protected] and [email protected]:0.95 of YOLO26n by 4.79 and 5.74 percentage points, respectively. On MSFFSD, the proposed method achieves 89.40% Precision, 83.12% Recall, 89.87% [email protected], and 69.76% [email protected]:0.95, while on D-Fire2 it obtains 76.25% Precision, 68.23% Recall, 72.88% [email protected], and 51.46% [email protected]:0.95. The ablation, comparative, class-specific, scale-aware, and visualization experiments further demonstrate the effectiveness of the proposed modules. With 3.04 M parameters, 6.0 GFLOPs, and an inference speed of 79.33 FPS, FSD-YOLO26 provides a lightweight solution for forest fire and smoke detection.

Authors

Institutions

Publication Details

Journal
Sensors
Published
2026-11-12
DOI
https://doi.org/10.3390/s26227139
Primary Topic
Fire Detection and Safety Systems
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

FSD-YOLO26: A Fire and Smoke Detection Method Based on YOLO26 for Forest Wildfire Detection

Carlos Ferrán, Dongming Liu, Dan Shan, Xuan Tong et al.
Sensors
Fire Detection and Safety Systems
article

FSD-YOLO26: A Fire and Smoke Detection Method Based on YOLO26 for Forest Wildfire Detection

Carlos Ferrán, Dongming Liu, Dan Shan, Xuan Tong, Zanqi Qiu
article en

Abstract

Forest fires pose a serious threat to ecological environments and public safety, while reliable fire and smoke detection remains challenging due to small target scales, diffuse smoke boundaries, complex vegetation backgrounds, and substantial variations in target appearance. This study aims to develop a lightweight and effective object detection framework for forest fire and smoke detection by enhancing fine-grained feature representation, spatial-semantic contextual modeling, and multi-scale feature interaction. To achieve this objective, FSD-YOLO26 is developed based on YOLO26n. A Fine-grained Feature Enhancement Module (FFEM) is introduced to strengthen the representation of small and fine-grained fire features. A Spatial-Semantic Context Enhancement Module (SSCEM) is designed to enhance spatial-semantic representation for diffuse and irregular smoke regions. Furthermore, a Dynamic Multi-scale Enhancement Fusion Module (DEMF) is proposed to improve multi-scale feature interaction and utilization under complex forest backgrounds. Under a consistent training protocol with 100 training epochs, a batch size of 10, and the AdamW optimizer with cosine learning-rate scheduling, the proposed method is evaluated on the FASDD, Multi-Scene Forest Fire-Smoke Dataset (MSFFSD), and D-Fire2 datasets using comparative, ablation, class-specific, scale-aware, and visualization experiments. On FASDD, FSD-YOLO26 achieves 77.74% Precision, 74.82% Recall, 83.23% [email protected], and 56.76% [email protected]:0.95, improving the corresponding [email protected] and [email protected]:0.95 of YOLO26n by 4.79 and 5.74 percentage points, respectively. On MSFFSD, the proposed method achieves 89.40% Precision, 83.12% Recall, 89.87% [email protected], and 69.76% [email protected]:0.95, while on D-Fire2 it obtains 76.25% Precision, 68.23% Recall, 72.88% [email protected], and 51.46% [email protected]:0.95. The ablation, comparative, class-specific, scale-aware, and visualization experiments further demonstrate the effectiveness of the proposed modules. With 3.04 M parameters, 6.0 GFLOPs, and an inference speed of 79.33 FPS, FSD-YOLO26 provides a lightweight solution for forest fire and smoke detection.

SensorsVol. 26(22)
Governors State University (US), Shenyang Jianzhu University (CN)
Department of Science and Technology of Liaoning Province
Fire Detection and Safety Systems
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon