This paper presents an AI-driven intelligent control system for precision spray applications that integrates real-time perception, neural network prediction, reinforcement learning optimization, and online Bayesian learning. A two-tier architecture is established: Tier 1 employs neural network-based feedback control for rapid pressure adjustment (0.1–0.3 MPa), while Tier 2 deploys an RL agent for optimal nozzle selection when pressure control is insufficient. The neural network model achieves high prediction accuracy with an RMSE of 3.2 $\mu$m on the test set. The RL agent demonstrates effective decision-making, attaining a 94.5% success rate in simulation and closed-loop experiments for maintaining target droplet size within $\pm$10 $\mu$m. Online Bayesian updates improve prediction accuracy by an average of 8.7% after every 100 field observations. Experimental validation demonstrates autonomous optimization, rapid convergence (3–5 cycles), and continuous learning without human retraining, offering a fully autonomous solution suitable for UAV deployment and precision agriculture.