BROWSE
Detail
Journal Article
Adaptive Progressive Learning for Real-time Fire Source Prediction with Limited Re-training Data
Cited 0 time in 
Share 
Share
- Authors
- Yusun Ahn, Soocheol Kim, Kyu Won Han, Kang Bok Lee, Hoe-Sung Yang, Kwangsoo Cho, Jin Hwa Ryu
- Issue Date
- 2026-02
- Citation
- Engineering Applications of Artificial Intelligence, v.166, no.A, pp.1-25
- ISSN
- 0952-1976
- Publisher
- Elsevier
- Language
- English
- Type
- Journal Article
- Abstract
- Fire poses a significant threat to life and property, making early detection paramount. False alarms compromise alarm system reliability, potentially delaying evacuations during actual fire events. This study introduces an artificial intelligence (AI)-driven adaptive learning framework using a convolutional neural network-long short-term memory (CNN-LSTM) model that compares and systematically analyzes fine-tuning and knowledge distillation (KD) to address these challenges. Progressive learning is employed to facilitate efficient re-training with minimal data. Performance analysis demonstrates that after an initial small-scale data collection, the base model yielded an F1-score (the harmonic mean of precision and recall) of 0.00 for both heptane and smoke check, underscoring its inability to classify new fire sources effectively. In contrast, fine-tuning achieved an F1-score of 0.87 for heptane and 0.93 for smoke check, demonstrating substantial improvement. Similarly, knowledge distillation attained F1-scores of 0.89 and 0.70, respectively. The proposed approach enhances the robustness and adaptability of fire detection systems, contributing to advanced safety measures in intelligent building environments by minimizing false alarms and enabling seamless data assimilation. By integrating adaptive learning and decision-making capabilities, the system operates in an autonomous manner, continuously managing and optimizing fire detection processes without human intervention.
- Keyword
- Adaptive progressive learning, Artificial intelligence, Fine-tuning, Fire detection, Knowledge distillation, Multi-angle wavelength
- KSP Keywords
- Adaptive learning, Convolution neural network(CNN), Data Collection, Decision-making, Detection Systems(IDS), Early detection, F1-score, Fine-tuning, Fire Detection, Fire source, Harmonic mean