EPFbased UAV reconnaissance moving target localization algorithm needs to use the EKF algorithm to calculate the mean and covariance of all particles in the sampling stage, which results in a large amount of computation. In this paper, an improved adaptive EPF algorithm based on KL divergence is proposed. The method uses the EKF algorithm to update the first half of the particles in the sampling phase. The latter half of the particles is still updated with the prior probability distribution, and then according to the KL divergence between the probability distributions of the two particle sets, the number of the particles at current moment is adaptively updated. Selecting the appropriate number of particles while ensuring accuracy greatly reduces the amount of calculation and improves the speed of operation. Through the verification with actually measured flight data, the average number of particles in each sampling period for this algorithm is 40, and the average calculation time in each sampling period is 8ms. Compared with EPF algorithm, this method can significantly reduce the calculation time while ensuring the positioning accuracy, and has certain engineering application value.