Image-based aerial fire detector based on cross-scale fusion

Zhang Pei; Ren Hengying; Tian Jiaqi; Chen Tong; Yan Weiwei; Zhang Wei

doi:10.12086/oee.2025.240253

Article navigation > Opto-Electronic Engineering > 2025 Vol. 52 > No. 1 > 240253

Next Article Previous Article

Zhang P, Ren H Y, Tian J Q, et al. Image-based aerial fire detector based on cross-scale fusion[J]. Opto-Electron Eng, 2025, 52(1): 240253. doi: 10.12086/oee.2025.240253

Citation:

Zhang P, Ren H Y, Tian J Q, et al. Image-based aerial fire detector based on cross-scale fusion[J]. Opto-Electron Eng, 2025, 52(1): 240253. doi: 10.12086/oee.2025.240253

Image-based aerial fire detector based on cross-scale fusion

1.
Electromechanical System Research Department, AVIC the First Aircraft Institute, Xi'an, Shaanxi 710089, China
2.
School of Microelectronics, Tianjin University, Tianjin 300072, China
3.
State Key Laboratory of Fire Science, University of Science and Technology of China, Hefei, Anhui 230027, China

Fund Project: National Science and Technology Major Project（J2019-VIII-0010-0171）

More Information

^*Corresponding author: renhengying1976@163.com
CSTR: 32245.14.oee.2025.240253

Received Date 27 October 2024

Revised Date 23 December 2024

Accepted Date 23 December 2024

Published Date 25 January 2025

Abstract

Abstract

Due to the low high air pressure during the flight, if a fire occurs in the cargo hold of the aircraft, the smoke particles are suspended in mid-air. The traditional smoke detector is difficult to detect, and there is also a high false alarm rate and difficult visualization in other environments, an image-based fire detector was designed, and the improved YOLOv5s algorithm was used to realize the pyrotechnic target detection. First, the backbone network is replaced with a lightweight GhostNet backbone network to facilitate hardware deployment. A collaborative attention module is embedded in the connection between the backbone and the converged network to strengthen the extraction of effective features. Then, according to the development and change characteristics of fire targets, the C3 structure in the feature fusion network was improved, the VoV-GSCSP module was built, and the Slim-ASFF module was embedded between the fusion network and the detection head, so as to jointly strengthen the feature fusion of different scales and realize the further lightweight of the overall network. Finally, the regression loss is replaced by focal EIOU, which solves the problem of penalty term failure and improves the prediction ability of positive samples. The image-based aviation fire detector takes the domestic AI chip RK3588 as the core, connects to the CMOS image sensor for data collection, and realizes information interaction with the airborne display system through the network. The test results show that the equipment can be arranged at the top four corners of the cargo compartment of the simulated aircraft, which can realize the flame alarm within 10 seconds and the smoke alarm within 20 seconds, which provides a feasible solution for ensuring the safety of the aircraft.
- RK3588 /
- fire-and-smoke detection /
- improvement of YOLOv5s /
- lightweight

FullText(HTML)

References

[1]	胡君健, 谢启源, 戎军, 等. 感烟火灾探测器的抗干扰实验及其分析[J]. 火灾科学, 2005, 14(2): 111−116. doi: 10.3969/j.issn.1004-5309.2005.02.009 CrossRef Google Scholar Hu J J, Xie Q Y, Rong J, et al. Experimental analysis on nuisance immunity of smoke detectors[J]. Fire Saf Sci, 2005, 14(2): 111−116. doi: 10.3969/j.issn.1004-5309.2005.02.009 CrossRef Google Scholar
[2]	Zhang R, Zhang W, Liu Y Y, et al. An efficient deep neural network with color-weighted loss for fire detection[J]. Multimed Tools Appl, 2022, 81(27): 39695−39713. doi: 10.1007/s11042-022-12861-9 CrossRef Google Scholar
[3]	高锴. 浅谈图像感烟探测产品在飞机货仓火灾探测中的应用[J]. 品牌与标准化, 2016, (7): 75−78. doi: 10.3969/j.issn.1674-4977.2016.07.006 CrossRef Google Scholar Gao K. Application of image smoke detection products in aircraft cargo compartment fire detection[J]. Brand Stand, 2016, (7): 75−78. doi: 10.3969/j.issn.1674-4977.2016.07.006 CrossRef Google Scholar
[4]	Blake D. Aircraft cargo compartment smoke detector alarm incidents on U. S. -registered aircraft, 1974–1999[R]. Atlantic City: Federal Aviation Administration, 2000. Google Scholar
[5]	何志祥, 王立纲, 孟超, 等. 基于多数据融合的复合火灾探测系统设计[J]. 消防科学与技术, 2019, 38(7): 977−980. doi: 10.3969/j.issn.1009-0029.2019.07.022 CrossRef Google Scholar He Z X, Wang L G, Meng C, et al. Design of composite fire detection system based on multi-data fusion[J]. Fire Sci Technol, 2019, 38(7): 977−980. doi: 10.3969/j.issn.1009-0029.2019.07.022 CrossRef Google Scholar
[6]	张红梅, 叶慧, 郑罡, 等. 多传感器飞机货舱火警探测系统研究[J]. 重庆理工大学学报(自然科学), 2017, 31(7): 176−181. doi: 10.3969/j.issn.1674-8425(z).2017.07.028 CrossRef Google Scholar Zhang H M, Ye H, Zheng G, et al. Research on the multi sensor fire detection system in aircraft cargo[J]. J Chongqing Univ Technol (Nat Sci), 2017, 31(7): 176−181. doi: 10.3969/j.issn.1674-8425(z).2017.07.028 CrossRef Google Scholar
[7]	何永勃, 张文杰, 杨伟, 等. 飞机货舱复合烟雾探测方法研究[J]. 中国安全科学学报, 2019, 29(1): 43−48. doi: 10.16265/j.cnki.issn1003-3033.2019.01.008 CrossRef Google Scholar He Y B, Zhang W J, Yang W, et al. Research on multi-sensor smoke detection method for aircraft cargo compartment[J]. China Saf Sci J, 2019, 29(1): 43−48. doi: 10.16265/j.cnki.issn1003-3033.2019.01.008 CrossRef Google Scholar
[8]	王哲. 飞机货舱防火设计要求研究[J]. 航空标准化与质量, 2014, (5): 13−15, 34. doi: 10.13237/j.cnki.asq.2014.05.004 CrossRef Google Scholar Wang Z. Research on fire protection design requirements for aircraft cargo compartments[J]. Aeronaut Stand Qual, 2014, (5): 13−15, 34. doi: 10.13237/j.cnki.asq.2014.05.004 CrossRef Google Scholar
[9]	张小雪, 王雨, 吴思远, 等. 基于级联稀疏查询机制的轻量化火灾检测算法[J]. 光电工程, 2023, 50(10): 230216. doi: 10.12086/oee.2023.230216 CrossRef Google Scholar Zhang X X, Wang Y, Wu S Y, et al. An improved lightweight fire detection algorithm based on cascade sparse query[J]. Opto-Electron Eng, 2023, 50(10): 230216. doi: 10.12086/oee.2023.230216 CrossRef Google Scholar
[10]	Han K, Wang Y H, Tian Q, et al. GhostNet: more features from cheap operations[C]//Proceedings of 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Seattle, 2020: 1577–1586. https://doi.org/10.1109/CVPR42600.2020.00165. Google Scholar
[11]	Zhang Y F, Ren W Q, Zhang Z, et al. Focal and efficient IOU loss for accurate bounding box regression[J]. Neurocomputing, 2022, 506: 146−157. doi: 10.1016/j.neucom.2022.07.042 CrossRef Google Scholar
[12]	Lin T Y, Goyal P, Girshick R, et al. Focal loss for dense object detection[C]//Proceedings of 2017 IEEE International Conference on Computer Vision, Venice, 2017: 2999–3007. https://doi.org/10.1109/ICCV.2017.324. Google Scholar
[13]	Hou Q B, Zhou D Q, Feng J S. Coordinate attention for efficient mobile network design[C]//Proceedings of 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Nashville, 2021: 13708–13717. https://doi.org/10.1109/CVPR46437.2021.01350. Google Scholar
[14]	Li H L, Li J, Wei H B, et al. Slim-neck by GSConv: a better design paradigm of detector architectures for autonomous vehicles[Z]. arXiv: 2206.02424, 2024. https://arxiv.org/abs/2206.02424v1. Google Scholar
[15]	Liu S T, Huang D, Wang Y H. Learning spatial fusion for single-shot object detection[Z]. arXiv: 1911.09516, 2019. https://doi.org/10.48550/arXiv.1911.09516. Google Scholar
[16]	Li C Y, Li L L, Jiang H L, et al. YOLOv6: a single-stage object detection framework for industrial applications[Z]. arXiv: 2209.02976, 2022. https://doi.org/10.48550/arXiv.2209.02976. Google Scholar
[17]	Wang C Y, Bochkovskiy A, Liao H Y M. YOLOv7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]//Proceedings of 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Vancouver, 2023: 7464–7475. https://doi.org/10.1109/CVPR52729.2023.00721. Google Scholar
[18]	Liu W, Anguelov D, Erhan D, et al. SSD: single shot MultiBox detector[C]//Proceedings of the 14th European Conference on Computer Vision, Amsterdam, 2016: 21–37. https://doi.org/10.1007/978-3-319-46448-0_2. Google Scholar
[19]	Tan M X, Pang R M, Le Q V. EfficientDet: scalable and efficient object detection[C]//Proceedings of 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Seattle, 2020: 10778–10787. https://doi.org/10.1109/CVPR42600.2020.01079. Google Scholar
[20]	Wang C Y, Ye I H, Liao H Y M. YOLOv9: learning what you want to learn using programmable gradient information[C]//Proceedings of the 18th European Conference on Computer Vision, Milan, 2024: 1–21. https://doi.org/10.1007/978-3-031-72751-1_1. Google Scholar
[21]	Wang A, Chen H, Liu L H, et al. YOLOv10: real-time end-to-end object detection[Z]. arXiv: 2405.14458, 2024. https://doi.org/10.48550/arXiv.2405.14458. Google Scholar
[22]	Zhao Y, Lv W Y, Xu S L, et al. DETrs beat YOLOs on real-time object detection[C]//Proceedings of 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Seattle, 2024: 16965–16974. https://doi.org/10.1109/CVPR52733.2024.01605. Google Scholar
[23]	中华人民共和国国家质量监督检验检疫总局, 中国国家标准化管理委员会. 特种火灾探测器: GB 15631–2008[S]. 北京: 中国标准出版社, 2009. Google Scholar General Administration of Quality Supervision, Inspection and Quarantine of the People’s Republic of China, Standardization Administration of the People's Republic of China. Special type fire detectors: GB 15631–2008[S]. Beijing: Standards Press of China, 2009. Google Scholar
[24]	闵芳. 盘点世界各国大型运输机[J]. 生命与灾害, 2013, 3: 34−37. https://cqvip.com/doc/journal/972803679 Google Scholar Min F. Overview of large transport aircraft in various countries[J]. Life Disaster, 2013, 3: 34−37. https://cqvip.com/doc/journal/972803679 Google Scholar

Overview

Overview

To address the challenges of traditional smoke detectors in identifying replaced smoke in the cargo hold of high-altitude, low-pressure aircraft, an innovative image-based fire-and-smoke detection system was developed using the domestic RK3588 embedded AI chip. This system employs an enhanced YOLOv5s detection algorithm tailored specifically for fire-and-smoke detection, incorporating several critical improvements to achieve high precision and operational efficiency. The backbone network of YOLOv5s is replaced with the lightweight GhostNet architecture, which significantly reduces computational requirements and the model’s parameter size, making it highly suitable for deployment on embedded devices with limited resources. To enhance feature extraction, a collaborative attention module is integrated between the backbone and the feature aggregation network, ensuring that critical features are captured effectively for better detection outcomes. In addition, the C3 structure in the feature fusion network is substituted with the VoV-GSCSP module. This modification not only enhances the integration of multi-scale features but also reduces computational complexity, enabling the system to handle high-resolution images more efficiently. To further optimize the system’s performance, the Slim-ASFF module is inserted between the feature fusion network and the detection head. This addition improves the combination of feature maps across varying scales, ensuring accurate detection of both small and large fire-and-smoke instances. The regression loss function is also updated by replacing the standard loss function with Focal EIOU. This improvement addresses challenges related to aspect ratio variations in the original loss function, enhancing the system’s ability to identify positive samples while reducing false alarms effectively. Experimental results on a self-constructed fire-and-smoke dataset demonstrate the system achieves a 2.0% increase in mean Average Precision at a 0.5 threshold (mAP50) and a 2.2% improvement at 0.5:0.95 thresholds (mAP50:95). These results demonstrate the algorithm’s effectiveness under challenging conditions, such as low light and high turbulence environments, making it highly reliable for real-world applications. The hardware of this system is centered around the RK3588 embedded processing board, which interfaces with a CMOS image sensor for real-time data acquisition. The processing board includes an RTSP streaming server, enabling the host computer to access the visual interface via the onboard LAN and an assigned IP address. Testing in a simulated cabin of 15 m × 8 m × 4 m demonstrated reliable performance, with flame alarms triggered within 10 seconds and smoke alarms within 20 seconds. All functional indicators met rigorous design specifications, confirming the system as a scalable, efficient, and reliable solution for fire-and-smoke detection in aircraft cargo holds. By combining advanced deep learning techniques, lightweight architectures, and optimized hardware, this system ensures compliance with the stringent demands of real-time monitoring in airborne environments.