Optimal Safety Zone Recommendation for Wildland Firefighters Using Deep Learning-Powered UAV-Based Forest Fire Detection

In firefighting operations, the conditions at the fireground can be complex and rapidly changing, with the spread of fire often occurring at an alarming rate. To address these challenges, it is crucial to establish wildland firefighter safety zone for firefighters in advance to respond to sudden changes in fire intensity. Typically, clearings in the forest, sparsely vegetated regions, and areas near ditches or ponds are designated as wildland firefighter safety zones. However, in extensive fire scenarios where terrain information is unclear, relying solely on manual searches for firefighter safety zones can be time-consuming, labor-intensive, and ultimately ineffective. This paper presents an optimal safety zone recommendation for wildland firefighters using UAV-based forest fire detection, integrating a road segmentation adapter and an object detection adapter. The method leverages UAVs for comprehensive detection along the fire perimeter, while the digital battlefield command and control platform facilitates three-dimensional reconstruction of the terrain and captures orthophoto image of the fire's outer edges. The image is transmitted to a ground artificial intelligence (AI) fusion gateway, where the road segmentation adapter extracts information about forest roads. To tackle the challenges posed by tree shadows that obscure road visibility, the object detection adapter first identifies the area of interest, followed by precise segmentation using the road segmentation adapter. The minimum inscribed circle algorithm is employed to calculate the narrowest width of each road, selecting those that meet vehicle passage requirements to form a candidate set. The object detection adapter further assesses whether safety zones surrounding each road in the candidate set meet specified area thresholds. Finally, by incorporating data on fire field weather, temperature, water source distribution, and slope, the method identifies optimal safety zone recommendation for wildland firefighters. Experimental results demonstrate that this approach can swiftly and accurately recommend optimal safety zone for large-scale fire incidents, offering greater efficiency and safety compared to traditional manual search methods. This represents a significant advancement toward minimizing firefighter casualties and enhancing overall firefighting effectiveness.