To address the vulnerability of UAV-based air-to-ground communication links to eavesdropping, this paper proposes a covert information-mapped generalized spatial and direction modulation (CIM-GSDM) system. In this system, information is concealed within the indices of activated receiver subsets and their selection combinations, while artificial noise orthogonal to the legitimate channel is introduced to disrupt potential eavesdroppers. To further enhance transmission security, a joint optimization framework for the precoding matrix and power allocation factor is developed, effectively managing multi-beam transmission and the distribution of artificial noise. The physical layer security metric, based on the system ’s secrecy rate, is derived and used as the optimization objective. To solve the resulting non-convex joint optimization problem, alternating optimization of the precoding matrix and power allocations factor is employed. A natural gradient descent method with Nesterov’s acceleration is proposed to efficiently update the precoding matrix, addressing computational complexity due to the large CIM-GSDM symbol candidate set. Furthermore, a suboptimal closed-form solution for the power allocation factor is derived based on maximizing the product of the legitimate user’s signal-to-noise ratio (SNR) and the eavesdropper’s interference-to-signal-plus-noise ratio (ISNR). Simulation results demonstrate that the proposed optimization algorithm significantly reduces the eavesdropper’s interception rate while ensuring the legitimate user’s achievable rate, effectively guaranteeing secure transmission in the CIM-GSDM system. Compared to traditional beamforming algorithms and fixed power allocation methods, the proposed algorithm achieves superior security performance.