Smartphone-Based Fuel Identification Model for Wildfire Risk Assessment Using YOLOv8

The increasing frequency and severity of wildfires, driven by climate change and rising greenhouse gas concentrations, makes the development of advanced wildfire risk assessment and early detection methods imperative. This study introduces an optimized deep learning model using YOLOv8 for fuel identification from roadside images, supporting the further development of Mobile Crowd Sensing (MCS) ecosystems for wildfire risk assessment. By leveraging mobile devices such as smartphones, this approach enables real-time estimation of vegetation density and dryness, crucial for wildfire forecasting paving the way for MCS schemes where devices can autonomously contribute to a geospatial vegetation database. Experimental results demonstrate the effectiveness of YOLOv8 on smartphones; specifically, the YOLOv8 Nano model, using half-precision, functions as a real-time object detector on the MediaTek Helio G99 processor. The model achieved a mean average precision (mAP) of 0.71 at mAP50 and 0.422 at mAP50-95, requiring only 6 megabytes of memory and averaging 220.3 ms per inference. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

Smartphone-Based Fuel Identification Model for Wildfire Risk Assessment Using YOLOv8 Chapter in Scopus