An advanced multi-source data fusion method utilizing deep learning techniques for fire detection

Fire poses a major risk to ecosystems, human life, and property. Due to the complexity of fire scenarios and the challenge of detecting small, dispersed targets, this paper proposes FireSmoke-YOLO, an innovative fire detection algorithm based on the You Only Look Once (YOLO) framework. Firstly, the Funnel Space Pyramid Pooling Fast (FSPPF) is proposed to replace the Spatial Pyramid Pooling Fast (SPPF) in the backbone module, enhancing the feature extraction of deep networks. Secondly, to improve the detection accuracy of small objects, a Small Object Detection Layer is integrated into the Neck module. Additionally, to better fuse deep and shallow features, a Dynamic Snake Convolution (DSC) is added after each concatenation layer in the Neck module. FireSmoke-YOLO was trained and evaluated with a custom fire dataset. Experimental results show FireSmoke-YOLO achieves 81. 4% a mean average precision at an intersection over union threshold of 0. 5 (mAP50) and 59% mAP50-95, surpassing the original model by 1. 9% and 8. 9%, respectively. The FSPPF layer reduces model size by 4. 4 megabytes, ensuring a lightweight design. The model shows strong generalization in processing satellite remote sensing data, validating its effectiveness in fire monitoring. This study enhances fire detection accuracy while providing an efficient, lightweight model for firefighting and environmental protection. FireSmoke-YOLO presents a new approach to fire monitoring technology, with broad application potential. Future work will focus on further optimization and expanding its use in diverse fire monitoring scenarios to improve public safety and ecological protection.

Authors

• Shikuan Wang
• Mengquan Wu
• Xinghua Wei
• Xiaodong Song
• Qingtong Wang
• Yanchun Jiang
• Jinkun Gao
• Lingyi Meng
• Zhipeng Chen School of Artificial Intelligence, Beijing University of Posts and Telecommunications, China
• Qiyue Zhang
• Yike Zhang
• Shuang Liang University of Electronic Science and Technology of China, Chengdu, China
• Shengtao Wei
• Longxing Liu
• Changbo Yi
• Jinyi Lv

Keywords

• Detection of fire and smoke
• Deep learning
• Object detection
• Computer vision
• Ecological protection