Korean Journal of Artificial Intelligence (한국인공지능학회지)

Korea Artificial Intelligence Association (한국인공지능학회)
권호 표지
DOI QR코드

DOI QR Code

Learnable Sobel Filter and Attention-based Deep Learning Framework for Early Forest Fire Detection

  • Sehun KIM (Defense Acquisition Program, Kwangwoon University) ;
  • Kyeongseok JANG (Department of Plasma Bio Display, Kwangwoon University) ;
  • Dongwoo LEE (Department of Plasma Bio Display, Kwangwoon University) ;
  • Seungwon CHO (Department of Holographic 3D Contents, Kwangwoon University) ;
  • Seunghyun LEE (Department of Ingenisum College Liberal Arts, Swangwoon University) ;
  • Kwangchul SON (Department of Electronics and Communications Engineering, Kwangwoon University)
  • Received : 2024.11.04
  • Accepted : 2024.12.05
  • Published : 2024.12.30
Download PDF
⟨ Previous Next ⟩

Abstract

Various techniques are being researched to effectively detect forest fires. Among them, techniques using object detection models can monitor forest fires over wide areas 24 hours a day. However, detecting forest fires early with traditional object detection models is a very challenging task. While they show decent accuracy for thick smoke and large fires, they show low accuracy for faint smoke and small fires, and frequently generate false positives for lights that are like fires. In this paper, to solve these problems, we focus on leveraging local characteristics such as contours and textures of fire and smoke, which are crucial for accurate detection. Based on this approach, we propose EDAM (Edge driven Attention Module) that performs enhancement by richly utilizing contour and texture information of fire and smoke. EDAM extracts important edge information to generate feature maps with emphasized contour and texture information, and based on this map, performs Attention Mechanism to emphasize key characteristics of smoke and fire. Through this mechanism, the overall model performance was improved, with APsincreasing from 0.154 to 0.204 and AP0.5 from 0.779 to 0.784, resulting in a significant improvement in APsvalue to 32.47%. In practice, the model applying this technique showed excellent inference speed while greatly improving detection performance for small objects compared to existing models and reduced false positive rates for building and street light illumination in nighttime environments that are easily mistaken for fire.

Keywords

Acknowledgement

This research was supported by Culture, Sports and Tourism R&D Program through the Korea Creative Content Agency(KOCCA) grant funded by the Ministry of Culture, Sports and Tourism(MCST) in 2024(Project Name: 3D holographic infotainment system design R&D professional human resources, Project Number: RS-2024-00401213, Contribution Rate: 100%)

References

  1. Al-Dabbagh, A. M., & Ilyas, M. (2023). Uni-temporal Sentinel-2 imagery for wildfire detection using deep learning semantic segmentation models. Geomatics, Natural Hazards and Risk, 14(1), 2196370. 
  2. Dong, H., Yuan, M., Wang, S., Zhang, L., Bao, W., Liu, Y., & Hu, Q. (2023). PHAM-YOLO: A parallel hybrid attention mechanism network for defect detection of meter in substation. Sensors, 23(13), 6052. 
  3. Fu, J., Zheng, H., & Mei, T. (2017). Look Closer to See Better: Recurrent Attention Convolutional Neural Network for Fine-Grained Image Recognition. In Proceedings of the 30th IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2017, Honolulu, HI, USA, 21-26 July 2017. 
  4. Jindal, P., Gupta, H., Pachauri, N., Sharma, V., & Verma, O. P. (2021). Real-time wildfire detection via image-based deep learning algorithm. In Soft Computing: Theories and Applications: Proceedings of SoCTA 2020, Volume 2 (pp. 539-550). Springer Singapore. 
  5. Liu, P., Xiang, P., & Lu, D. (2023). A new multi-sensor fire detection method based on LSTM networks with environmental information fusion. Neural Computing and Applications, 35(36), 25275-25289. 
  6. Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., ... & Chintala, S. (2019). Pytorch: An imperative style, high-performance deep learning library. Advances in neural information processing systems, 32. 
  7. Redmon, J. (2016). You only look once: Unified, real-time object detection. Proceedings of the IEEE conference on computer vision and pattern recognition. 
  8. Ripple, W. J., Wolf, C., Gregg, J. W., Rockstrom, J., Mann, M. E., Oreskes, N., ... & Crowther, T. W. (2024). The 2024 state of the climate report: Perilous times on planet Earth. BioScience, biae087. 
  9. Scott, J. H., Thompson, M. P., & Calkin, D. E. (2013). A wildfire risk assessment framework for land and resource management.
  10. Vaswani, A. (2017). Attention is all you need. Advances in Neural Information Processing Systems. 
  11. Wang, C. Y., Bochkovskiy, A., & Liao, H. Y. M. (2023). YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition (pp. 7464-7475). 
  12. Wang, C. Y., Liao, H. Y. M., & Yeh, I. H. (2022). Designing network design strategies through gradient path analysis. arXiv preprint arXiv:2211.04800.
  13. Wang, F., Jiang, M., Qian, C., Yang, S., Li, C., Zhang, H., ... & Tang, X. (2017). Residual attention network for image classification. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 3156-3164). 
  14. Woo, S., Park, J., Lee, J. Y., & Kweon, I. S. (2018). Cbam: Convolutional block attention module. In Proceedings of the European conference on computer vision (ECCV) (pp. 3-19). 
  15. Yu, L., Wang, N., & Meng, X. (2005, September). Real-time forest fire detection with wireless sensor networks. In Proceedings. 2005 International Conference on Wireless Communications, Networking and Mobile Computing, 2005. (Vol. 2, pp. 1214-1217). IEEE. 
  16. Zope, V., Dadlani, T., Matai, A., Tembhurnikar, P., & Kalani, R. (2020, May). IoT sensor and deep neural network based wildfire prediction system. In 2020 4th International Conference on Intelligent Computing and Control Systems (ICICCS) (pp. 205-208). IEEE.