The broadcast nature of wireless channels exposes them to severe eavesdropping risks, particularly when the eavesdropper is geographically close to the legitimate receiver. In such scenarios, the Channel State Information (CSI) of both parties exhibits high correlation or even superior quality for the eavesdropper, rendering traditional power optimization methods ineffective in reducing the eavesdropping rate. To address this issue, a secure transmission strategy for unmanned aerial vehicle relays is proposed, based on power allocation factor optimization. By dynamically adjusting the power allocation factor to control the power ratio of Artificial Noise (AN), adaptive interference control over the eavesdropping channel is achieved. A four-node communication system is constructed, comprising a source node, a UAV relay, a legitimate receiver, and an eavesdropper. A non-convex optimization problem is formulated with the objective of maximizing secrecy capacity. Through the introduction of slack variables and the Successive Convex Approximation (SCA) method, the original problem is transformed into a convex optimization problem. Combined with the Staircase Water Filling (SWF) method, a closed-form analytical solution for the power allocation factor is derived, and an alternating iterative optimization algorithm is proposed to achieve global parameter optimization. Simulation results demonstrate that the proposed algorithm reduces eavesdropping capacity by 76% compared to traditional power optimization algorithms, significantly enhancing the system's security performance.