Adaptive Multispectral Perception and Distributed Trajectory Optimization for Low Altitude Economic UAV Swarms Operating in Complex Mountainous Agricultural Ecosystems of Yunnan

Abstract

The emergence of low altitude economic ecosystems presents transformative potential for precision agriculture in topographically challenging regions. This study introduces an integrated framework designated Adaptive Multispectral Edge Fusion Trajectory Optimization for low altitude unmanned aerial vehicle swarms deployed in the mountainous tea plantations of Yunnan Province. The proposed architecture unifies an adaptive middleware stack for resilient inter vehicle communication, a multispectral machine vision network with cross channel spectral attention, and a distributed model predictive control trajectory planner that explicitly encodes orographic constraints, wind drift, and energy budgets. A digital twin enabled hardware in the loop simulation environment calibrated with field data from Pu'er and Dali operational sites serves as the primary validation platform. Comparative assessments against four baseline perception pipelines, three classical trajectory planners, and two conventional middleware stacks demonstrate that the proposed system achieves a mean average precision of 0.893 for tea plant stress detection across five spectral bands, reduces trajectory length by 18.4 percent in high complexity gorges, and maintains sub 35 millisecond end to end latency under 30 percent packet loss conditions. These findings illustrate that coupling adaptive software architectures with multispectral machine vision and distributed planning significantly enhances operational safety, chemical use efficiency, and mission endurance in rugged mountainous agricultural environments. Notwithstanding these advances, prolonged adverse weather interactions and full beyond visual line of sight regulatory compliance require further validation.