UAV-Assisted AR Visualization for Real-Time Agricultural Field Monitoring Using Multispectral Data and U-Net Segmentation

The requirement for UAVs in dynamic and uncertain application environments, such as infrastructure inspection, disaster management, and delivery operations, has become more apparent. While existing path planning methods, such as Dijkstra and PRM, work reasonably well in static environments, they present considerable difficulties when dealing with dynamic obstacles, shifting priorities, and real-time environmental dynamics. High computational overhead, slower re-optimization, and lower adaptability, which reduce mission efficiency and success rates, are common to these methods. To overcome these challenges, this research presents a novel Adaptive Path Planning Framework, which combines real-time obstacle detection, AR-assisted visualization, and finally, the A* algorithm for optimizing a least travel time path. Secondly, the paper presents a proposed system that allows UAVs to update their navigation paths in an adaptive, on-demand manner, dynamically switching to alternative paths in the event of newly detected obstacles or in response to priority shifts, with minimal computational overhead. The paper quantitatively evaluates that the proposed framework results in a 95% collision avoidance success rate in high-density obstacle scenarios while reducing path deviations by 30% compared to current approaches. Additionally, the system achieved an average re-optimization time of 0.1 seconds, some orders of magnitude faster than traditional methods, along with greater adaptability. On this data, the results showed that the proposed framework is robust and efficient in real-world scenarios, which makes it suitable for application with dynamic re-planning and precision navigation. This work provides new insights by coping with the limitations of the current methods and contributes to promoting advances in UAV navigation systems with a solid groundwork in multi-UAV systems, energy-efficient path planning, and machine learning-driven predictive capabilities.