Pixel-wise Active Fire Classification in Landsat-8 using Deep Learning

This repository contains the code to perform supervised classification of active fire pixels from Landsat-8 images.

Requirements

Please see requirements.txt

Usage

• dataset.py: contains the ActiveFire dataset instance
• inference.py: runs inference on the test set and saves the results
• main.py: runs training
• model.py: defines the model
• trainer.py: defines the train/val/test loops

Simply download the dataset and accordingly revise the paths in main.py.

Dataset

The dataset used in this study: Active fire detection in Landsat-8 imagery: A large-scale dataset and a deep-learning study

Sample Results