The model is based on a pre-trained YOLOv5 Object Detection model, which was finetuned on a custom made dataset.

The pipeline was then encapsulated in a FastAPI-based REST API, which was then deployed on Heroku.

The original plan was to utilize a drone, with the model and the pipeline mounted on top, to discover possible wildfires.

Wildfire detection overview

Project repository

 

Providing state of the art solutions for fire and smoke detection.

At the current stage, it's able to provide:

  • Fire & smoke detection: A custom object detection and image classification model able to identify the presence of smoke and fire in a multitude of scenarios, including forests.