A Two-Stage UAV Team Path-Planning Framework for Wildfire Search Employing Clustering and Self-Organizing Map Neural Networks

This study introduces a novel two-stage framework for optimizing the path planning of heterogeneous UAV teams in wildfire search missions across expansive, complex environments. The first stage efficiently generates potential waypoints from a probabilistic wildfire risk map by employing a combination of thresholding, density-based clustering, and k-means clustering. The second stage utilizes a self-organizing map (SOM) neural network to generate UAV trajectories that maximize the cumulative probability of target detection. Extensive simulations using spatial datasets of wildfire risk verify the efficacy of the proposed framework. Comparative analyses with widely used heuristic optimization approaches reveal that the SOM method consistently produces more efficient paths with higher cumulative detection probabilities. Moreover, the SOM-based strategy achieves superior path smoothness and significantly reduces computational overhead, making it particularly well-suited for large-scale search missions involving diverse UAVs.