The communication efficiency of unmanned aerial systems directly determines their ability to accomplish combat missions. Traditional dynamic efficiency evaluations use probabilistic finite state machines to characterize communication efficiency. However, as the number of collaborative devices increases, computational complexity and fluctuations in performance states rise significantly, compromising the robustness and effectiveness of the evaluation process. Therefore, an improved Kalman filtering approach is proposed, which utilizes the Mahalanobis distance classifier to identify the linear and anomalous components within the communication system's performance states, enabling the recognition of the system's status. Subsequently, Kalman filtering is employed to model the communication efficiency under different states separately and conduct corresponding dynamic assessments, thereby balancing modeling accuracy and computational efficiency. Taking unmanned aerial vehicle communication as an example, a communication efficiency evaluation system is established. The results are used to optimize the efficiency indicators, distribution, and reliability of each node, verifying the effectiveness of the method and achieving precise improvement in the communication efficiency of unmanned aerial vehicle networks.