Paper link: Using knowledge inference to suppress the lamp disturbance for fire detection
Our project is developed based on Firenet. Please follow the official Firenet README.md for installation and preparation.
conda create -n fire python=3.7
conda activate fire
pip install tensorflow=1.15.0
pip install -r requirements.txt
python demo.py --save-txt --savetxt --FLremove
The results of the lamp and fire detection (suppression is not showed) is in the output folder. The file result.txt shows the fire probability.
python demo.py --save-txt --savetxt --saveimg --FLremove
The results of the lamp and fire detection (Rectangular mask) is in the output folder.
python demo.py --save-txt --savetxt --saveimg --grabcut --FLremove
The results of the lamp and fire detection (Segmentation mask) is in the output folder.
1.Download fire data from https://collections.durham.ac.uk/files/r2d217qp536#.X5F5G2gzZnK
2.Download lamp data from openimageV4：
git clone https://github.com/EscVM/OIDv4_ToolKit.git
following its instruction, then use the following command to get lamp data:
python3 main.py downloader --classes Lamp --type_csv train(test,validation)
Put all the pictures and labels in a same folder(data/JPEGImages)
If you want to train Fire-only weights:
python3 train.py --epochs n(epochs you want) --weights weights/yolov3.pt --cfg cfg/yolov3-fireonly.cfg --data data/fire.data --name data/fire.names --single-cls
If you want to train Fire-lamp weights:
python3 train.py --epochs n(epochs you want) --weights weights/yolov3.pt --cfg cfg/yolov3.cfg --data data/fires.data --name data/fires.names
@article{SUN2021124,
author = {Kailai Sun and Qianchuan Zhao and Xinwei Wang},
title = {Using knowledge inference to suppress the lamp disturbance for fire detection},
journal = {Journal of Safety Science and Resilience},
volume = {2},
number = {3},
pages = {124-130},
year = {2021},
issn = {2666-4496},
doi = {https://doi.org/10.1016/j.jnlssr.2021.07.002},
url = {https://www.sciencedirect.com/science/article/pii/S266644962100027X},
}