Evaluation of Machine Learning Methods for Fire Risk Assessment from Satellite Imagery

• Type: Article in International Conference Proceedings Book
• Date: 2025

Authors

Bittencourt, JCN (Author) Other; Costa, DG (Author) FEUP; Paulo Portugal (Author) FEUP; Francisco Vasques (Author) FEUP

Conference proceedings International

Title: PROGRESS IN ARTIFICIAL INTELLIGENCE, EPIA 2024, PT I

Pages: 398-409

23rd EPIA Conference on Artificial Intelligence-EPIA

Viana do Castelo, PORTUGAL, SEP 03-06, 2024

Indexing

ISI Web of Knowledge - 0 Citations

Scopus - 0 Citations

Other information

Authenticus ID: P-017-CTD

DOI: 10.1007/978-3-031-73497-7_32

Abstract (EN): Recognising the critical role forests play in global biodiversity and the increasing threat of wildfires, this work exploits advanced geoscientific technologies and machine learning techniques to improve fire risk prediction and management. The primary objective is to develop a Convolutional Neural Network (CNN) that maps remotely sensed images to fire risk levels using a refined subset of the FireRisk dataset. The employed dataset contains 7,644 images categorised into five fire risk classes. Based on it, this work benchmarks the performance of InceptionResNetV2 and Vision Transformer models, which have been pre-trained on extensive datasets and fine-tuned for fire risk classification. The achieved custom CNN model achieves an accuracy and F1 score of 72%, demonstrating its potential for this application.

Language: English

Type (Professor's evaluation): Scientific

No. of pages: 12