A Hybrid Deep Learning Model for Early Forest Fire Detection DOI Open Access

Akhror Mamadmurodov,

Sabina Umirzakova, Mekhriddin Rakhimov

et al.

Forests, Journal Year: 2025, Volume and Issue: 16(5), P. 863 - 863

Published: May 21, 2025

Forest fires pose an escalating global threat, severely impacting ecosystems, public health, and economies. Timely detection, especially during early stages, is critical for effective intervention. In this study, we propose a novel deep learning-based framework that augments the YOLOv4 object detection architecture with modified EfficientNetV2 backbone Efficient Channel Attention (ECA) modules. The substitution leverages compound scaling Fused-MBConv/MBConv blocks to improve representational efficiency, while lightweight ECA enhance inter-channel dependency modeling without incurring significant computational overhead. Additionally, introduce domain-specific preprocessing pipeline employing Canny edge CLAHE + Jet transformation, pseudo-NDVI mapping fire-specific visual cues in complex natural environments. Experimental evaluation on hybrid dataset of forest fire images video frames demonstrates substantial performance gains over baseline contemporary YOLO variants (YOLOv5–YOLOv9), proposed model achieving 97.01% precision, 95.14% recall, 93.13% mAP, 92.78% F1-score. Furthermore, our outperforms fourteen state-of-the-art approaches across standard metrics, confirming its efficacy, generalizability, suitability real-time deployment UAV-based computing platforms. These findings highlight synergy between architectural optimization domain-aware high-accuracy, low-latency wildfire systems.

Language: Английский

A Hybrid Deep Learning Model for Early Forest Fire Detection DOI Open Access

Akhror Mamadmurodov,

Sabina Umirzakova, Mekhriddin Rakhimov

et al.

Forests, Journal Year: 2025, Volume and Issue: 16(5), P. 863 - 863

Published: May 21, 2025

Forest fires pose an escalating global threat, severely impacting ecosystems, public health, and economies. Timely detection, especially during early stages, is critical for effective intervention. In this study, we propose a novel deep learning-based framework that augments the YOLOv4 object detection architecture with modified EfficientNetV2 backbone Efficient Channel Attention (ECA) modules. The substitution leverages compound scaling Fused-MBConv/MBConv blocks to improve representational efficiency, while lightweight ECA enhance inter-channel dependency modeling without incurring significant computational overhead. Additionally, introduce domain-specific preprocessing pipeline employing Canny edge CLAHE + Jet transformation, pseudo-NDVI mapping fire-specific visual cues in complex natural environments. Experimental evaluation on hybrid dataset of forest fire images video frames demonstrates substantial performance gains over baseline contemporary YOLO variants (YOLOv5–YOLOv9), proposed model achieving 97.01% precision, 95.14% recall, 93.13% mAP, 92.78% F1-score. Furthermore, our outperforms fourteen state-of-the-art approaches across standard metrics, confirming its efficacy, generalizability, suitability real-time deployment UAV-based computing platforms. These findings highlight synergy between architectural optimization domain-aware high-accuracy, low-latency wildfire systems.

Language: Английский

Citations

0