The intent for this project

Build a Wildfire detection model that drones can use to pre-emptively find wildfires before they really go 'wild'.

#1 Finding the right dataset

• There was a discussion between going the satellite or drone images route. We decided to stick to drone fire images since satellite images often have refresh times in the days. By the time we detect the fires, it would be too late.
• We decided to use a dataset of controlled wildfires from Northern Arizona USA after a general internet search on what was available.
  • Link: https://ieee-dataport.org/open-access/flame-dataset-aerial-imagery-pile-burn-detection-using-drones-uavs#files
  • To simplify our task, we focused on image data instead of videos - so we downloaded files #7 & #8 from the dataset.
  • Note that you need to make an account to download the datasets, but downloading them in and of themselves is free.

#2 EDA for Wildfire Images

• Everyone did their own EDA.
• My EDA was focused on the 3 questions:
  • Q1: How many images are in the train & test sets with & without fire? [To check for class imbalance]
  • Q2: The pixel sizes of the images in each? [If different sizes, would have to crop before modelling]
  • Q3: How are the pictures labelled? [To check if I can rely on it]

Find the code & my finding in my EDA.

EDA.ipynb

#3 Divvying up data for labeling