Adaptive Coverage Planning at No-Spray Buffer Zone Boundaries for UAV Operations

UAV-based coverage missions near exclusion zones—such as no-spray buffer zones in precision agriculture (watercourse margins, residential setbacks, organic field boundaries, apiary protection areas, and power line corridors), urban no-fly zones, and post-disaster cordons—demand coverage path planning (CPP) that maximizes spray coverage while strictly avoiding regulatory violations. These buffer zones are typically delineated as rectangular exclusion regions, creating abrupt obstacle edges where conventional uniform grid decomposition faces a fundamental dilemma: coarse grids leave untreated margins that become pest and disease reservoirs, while fine grids incur prohibitive computational cost. Furthermore, deadlock situations frequently arise in fields fragmented by multiple buffer zones, and naive A*-based escape maneuvers waste flight time without contributing spray coverage. This paper proposes an adaptive CPP framework integrating environment modeling, edge refinement, and deadlock escape into a unified three-level optimization. First, a spray-swath-driven adaptive grid decomposition dynamically couples grid resolution with the spray swath width, enabling efficient coverage of large open areas while preserving fine resolution near buffer zone boundaries. Second, a multi-level threshold hierarchical edge refinement strategy classifies boundary-adjacent void regions by size, applying targeted path interpolation, local re-planning, and region-level CPP to achieve boundary-tight coverage without encroaching into exclusion zones. Third, an improved A* deadlock escape algorithm selects the escape target by a goal-evaluation cost combining a Euclidean distance term and a binary obstacle-adjacency indicator, guiding the UAV along escape trajectories that advance the spray mission. Simulation experiments on four representative farmland scenarios demonstrate that the proposed method achieves 100% coverage in all test cases while reducing path length by approximately 48% and scan repetition by over 70% compared with fixed-grid baselines. These results are obtained under idealized simulation conditions and have not yet been validated in field trials; operational benefits such as pesticide savings are, therefore, estimated under the specified assumptions.