This study addresses the real-time wind field estimation requirements for energy-saving autonomous flight of small, low-cost unmanned aerial vehicles (UAVs) operating in complex wind environments. By integrating an aerodynamic model into an extended Kalman filter (EKF) estimation algorithm, both the flight state variables and wind field information are estimated. The application of this filtering algorithm effectively mitigates the high-frequency noise issues inherent in essential onboard sensors (such as global navigation satellite systems, inertial navigation systems, air data systems). Moreover, the EKF estimation algorithm incorporating the aerodynamic model demonstrates high accuracy—with errors within 0.2 m/s—and rapid convergence, enabling efficient estimation of both steady and various typical time-varying wind fields. Its low computational complexity further supports effective onboard real-time computation. The study also analyzes scenarios involving failure or absence of angle-of-attack and angle of sideslip sensors, exhibiting strong robustness. Additionally, by accounting for aerodynamic model errors, a quantitative analysis shows that within a 10% estimation error range of the lift coefficient, the wind speed estimation accuracy is approximately 0.3 m/s, thereby providing a reliable validation of wind speed estimates under conditions of model inaccuracy. Overall, this research establishes an environmental perception technology foundation for autonomous decision-making and control in energy-saving UAV flight, constituting an important technical guarantee for achieving embodied intelligence.