ConvLSTM-BASED HYBRID MODEL FOR FOREST FIRE IMAGE CLASSIFICATION IN VIETNAM
Keywords
DOI:
https://doi.org/10.34238/tnu-jst.15980Abstract
Forest fires represent a critical issue in Vietnam, particularly during the dry season. This study proposes the utilization of the ConvLSTM neural network for the early detection of forest fires from imagery data. Unlike traditional methods that process individual frames in isolation, ConvLSTM is capable of combining spatial feature extraction (via convolutional layers) with temporal dependency modeling (via Long Short-Term Memory). We utilize a publicly available dataset comprising 999 images with and without fire. To apply ConvLSTM to static images, time-series sequences are generated using a sliding window approach across data augmentation variants. Experimental results demonstrate that the model achieves an accuracy of 98.3%, a precision of 98.1%, a recall of 98.6%, and an F1-score of 98.2%, significantly outperforming standalone CNN or LSTM models. However, the study is constrained by its limited dataset size, which exhibits an imbalance where 75% of the samples contain fire, as well as a lack of Vietnam-specific data. Future research lines should focus on expanding to larger datasets that better reflect local conditions and deploying the system in real-time on edge devices.Downloads
References
A. B. Abdusalomov et al., "An improved forest fire detection method based on Detectron2," Sensors, vol. 23, no. 3, 2023, Art. no. 1512.
J. Mei and W. Zhang, "Early fire detection algorithm based on ViBe and machine learning," Acta Opt. Sin., vol. 38, 2018, Art. no. 0710001.
K. Dimitropoulos, P. Barmpoutis, and N. Grammalidis, "Higher order linear dynamical systems for smoke detection," IEEE Trans. Circuits Syst. Video Technol., vol. 27, pp. 1143–1154, 2017.
J. Zhang et al., "ATT squeeze U-net: A lightweight network for forest fire detection and recognition," IEEE Access, vol. 9, pp. 10858–10870, 2021.
R. Ghosh and A. Kumar, "A hybrid deep learning model combining CNN and RNN to detect forest fire," Multimedia Tools Appl., vol. 81, pp. 38643–38660, 2022.
A. Larsen et al., "A deep learning approach to identify smoke plumes in satellite imagery in near-real time," J. Expos. Sci. Environ. Epidemiol., vol. 31, pp. 170–176, 2021.
X. Yin, M. Bai, Y. Peng, B. Chen, H. Lai, W. Kou, et al., “LSTM neural network integrating remote sensing and meteorological data reveals forest fire risk trajectories in China over the past 100 years,” Agricultural and Forest Meteorology, vol. 375, 2025, Art. no. 110884.
C. Mambile, S. Kaijage, and J. Leo, "Deep learning models for enhanced forest-fire prediction at Mount Kilimanjaro," Natural Hazards Research, vol. 5, no. 2, pp. 335–347, 2025.
H. V. Le, D. A. Hoang, and G. T. Tran, “Optimizing machine learning models for enhanced forest fire susceptibility mapping in Gia Lai province,” Journal of Mining and Earth Sciences, vol. 66, no. 2, pp. 2-14, 2025.
V. T. Pham, T. A. T. Do, and H. D. Tran, “Classifying forest cover and mapping forest fire susceptibility in Dak Nong province, Vietnam utilizing remote sensing and machine learning,” Ecological Informatics, vol. 79, 2024, Art. no. 102392, doi: 10.1016/j.ecoinf.2023.102392.
T. N. A. Do, M. K. Hoang, and A. T. Le, “Improvement the fire detection performance using computer vision based on YOLOV5 neural network,” Hanoi University of Industry Journal of Science and Technology, vol. 5, pp. 48-53, 2022.
A. Saied et al., “FIRE Dataset,” Kaggle, 2018. [Online]. Available: https://www.kaggle.com/ datasets/phylake1337/fire-dataset. [Accessed Aug. 07, 2025].
