Fire-Detection

This is a non-deep learning fire detection pipeline inspired by this paper. Our method comprises of three parts: color space classifier, color component classifier and texture classifier. Our models were trained and tested on the BoWFire Dataset and is able to detect fire from static images with an accuracy of 80%.

Pipeline

Environment

Clone the code:

git clone https://github.com/Lukeli0425/Fire-Detection.git

Install the required packages for this repo:

pip install -r requirements.txt

Train the Models

Train the models with

python train.py

The trained models will be saved under the models folder.

Test the Models

Test the models on the BoWFire Dataset with

python test.py

The default setting is to use all three classifiers together. The results will be saved under the resultsfolder. If you want to use certain classifiers, run:

python test.py  --color_space [True/False]  --color_component [True/False]  --texture [True/False]

Train and Test together

chmod +x run.sh
./run.sh

Best Results

File list

...
+ BoWFireDataset              # train & test dataset
+ models                      # saved models
+ references                  # reference paper
+ results                     # experiment results
README.md 
requirements.txt              # environment prerequisites
colorspace.py                 # color space classifier
component.py                  # color component classifier
texture.py                    # texture classifier
train.py                      # train the model
test.py                       # test the model
run.sh                        # train and test the model together