A Fire Detection and Assessment Method based on YOLOv8

Fires annually result in significant loss of life and economic damage, with a substantial portion of these losses attributed to a lack of understanding regarding fire types and their respective environments. Among all fire types, forest and grassland fires stand out as the most severe disasters due to their extensive scope and potential for rapid spread. Therefore, flame recognition and disaster assessment serve as pivotal methods for detecting fires and acquiring fire-related information in complex scenarios. This study employs diverse YOLOv8 models trained on a custom dataset to evaluate their classification performance and training speed or inference time. Furthermore, a fire risk assessment model is introduced specifically for forest and grassland fires. The findings indicate that the YOLOv8n model with the smallest network architecture is optimal for deployment on lightweight machinery, demonstrating superior performance in fire detection and classification efficiency and accuracy. Leveraging environmental parameters detected by the fire risk assessment model, swift and precise evaluations of the severity of forest and grassland fires are attainable. In the future, this model holds promise for applications in lightweight contexts such as fire detection, fire alerts, and risk assessment.