Teacher: Giovanna Venuti
Tutor: Dr. Daniela Stroppiana – CNR/IREA
Description: Accurate mapping of burned areas is essential for monitoring wildfire impacts and supporting environmental management. High-resolution satellite imagery provides detailed information on fire-affected areas, but manual interpretation is time-consuming and difficult to scale.
The objective of this project is to implement a deep learning model for burned area mapping using fire perimeter data derived from PlanetScope imagery (this dataset is already available). A deep learning model for image segmentation will be implemented to classify each pixel as burned or unburned. The model will be trained using satellite image patches as input and burned area masks derived from fire perimeters as target labels.
The project will focus on developing scripts for data preparation, model training, and testing using data from multiple fire events across different global sites for the year 2023. The project will also include script to support the analysis of results from the testing phase as a function of the biomes, geographical area, season and number of fires.
Technologies: The implementation will be carried out in Python or R, using appropriate deep learning libraries.