Early Wildfire Detection from Images

A project for detection and classification of wildfire from images

This project has two repositories. One for classification and one for wildfire detection using bounding boxes. Fire detection using yolov3 is from the forked version of https://github.com/experiencor/keras-yolo3 along with some modifications

---

SETUP

Prerequisites:

Python3

1. Clone the following repos

    git clone https://github.com/pravrc/wildfireDetection.git
    git clone https://github.com/pravrc/yolov3.git 

2. Install packages

    pip install -r requirements.txt

or

    pip install fastai
    pip install tensorflow
    pip install opencv-contrib-python

---

DATASETS

Datasets for training and testing can be downloaded from S3 using wget

    wget http://pravrc-wildfire.s3.amazonaws.com/wildfiredata.zip
    unzip wildfiredata.zip

wildfire_train_data and wildfire_test_data are the folders containing training and test data respectively

---

TRAINING

1. Classification:

Make sure you are in ~/wildfireDetection. Make sure you setup all the training configuration in file ~/wildfireDetection/configs/trainClassifier.json. The existing sample has preset values which can be edited

    cd ~/wildfireDetection
    python wildfireDetection/trainClassifier.py -c configs/trainClassifier.json

2.Detection:

Edit config.json appropriately

    cd ~/keras-yolo3
    python train.py -c config.json

---

INFERENCE

1. Classification

Make sure you setup all the inference configuration in file ~/wildfireDetection/configs/inferenceClassifier.json. The existing sample has preset values which can be edited

    cd ~/wildfireDetection
    python wildfireDetection/inferenceClassifier.py -c configs/inferenceClassifier.json    

2. Detection

Edit config.json to set parameters

    cd ~/keras-yolo3
    python predict.py -c config.json -i INPUT_FILE

---