Year: 2025
Type: dissertation
Abstract: Wildfires are becoming increasingly frequent and destructive, posing significant risks to ecosystems, human health, and infrastructure. Timely and accurate detection of wildfire smoke is crucial for early intervention, which can mitigate the devastating effects of these fires. This research explores the use of Unmanned Aerial Vehicles (UAVs) equipped with optical sensors and deep learning models, specifically YOLOv8, to detect wildfire smoke in challenging environments. The primary objective of this study is to evaluate whether optical sensors, when paired with YOLOv8, can provide a viable and cost-effective alternative to traditional thermal sensors for early smoke detection, particularly in areas with dense vegetation, rugged terrain, and varying environmental conditions. Field experiments were conducted in the Büelenwald region of Switzerland, where controlled wildfire scenarios were simulated to generate diverse smoke conditions. The YOLOv8 model, known for its real-time object detection capabilities, was trained and tested on this specialized dataset to assess its performance in detecting smoke at early stages. The study focuses on optimizing detection accuracy through various techniques, including multi-rotation image analysis, area thresholding, and fine-tuning of the detection algorithm. The findings suggest that optical sensors equipped with YOLOv8 can reliably detect smoke, offering significant advantages in terms of operational feasibility, speed, and cost-effectiveness compared to traditional thermal sensors. However, the study also highlights the limitations of optical sensors under certain conditions, such as low visibility and adverse weather. By examining the influence of environmental factors like weather, terrain, and vegetation on detection performance, the research provides valuable insights into improving the reliability of smoke detection systems. This research contributes to the development of advanced, real-time wildfire monitoring solutions, demonstrating the potential of UAVs, deep learning, and optical sensors in enhancing wildfire detection and management. The results offer a pathway for future applications in wildfire monitoring, providing scalable and adaptive detection strategies that can help mitigate the growing risks posed by wildfires.
DOI: 10.5167/uzh-278951
Author: Capitanio, Luca
Language: English
Subfield: Safety, Risk, Reliability and Quality
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W6929944660
Cites:
Cited by:
Related to: