To address the problem of poor cooperative navigation of master-slave UAV swarms in complex environments with time-varying non-Gaussian noise due to external interference， an improved Particle Filter （PF） algorithm is proposed to improve navigation accuracy and reduce the accuracy requirements of the slave measurement equipment. Firstly， the observation model is established by taking the high-precision Inertial Navigation System （INS） and Global Position System （GPS） navigation information from the host machine as the reference and combining it with the low-precision sensors on the slave machine. Then， the fusion of multi-source navigation information is achieved using an improved PF algorithm. For the importance probability density function selection and particle degradation problems of PF， the Levenberg-Marquardt iterative method is introduced on the basis of Extended Particle Filter （EPF） to ensure the stability and convergence of the filter. In the resampling phase， the fast resampling method is proposed to classify the obtained particle set， and the medium weight particles are not resampled anymore， while the rest of the particles are optimized in the normalization process using adaptive weight factors to make the obtained sample particle weights more uniform， thus improving the computational efficiency and navigation real-time. The simulation results show that compared with other PFs， the proposed algorithm can effectively improve the accuracy of Unmanned Aerial Vehicle （UAV） formation cooperative navigation and has certain practical value.