To deal with the problem of excessive amount of computation in the collaborative positioning and navigation of UAV swarm in complex unknown environment this paper proposes two DRL methods for collaborative information screening of UAV swarm.The multi-UAV localization problem is mapped to a Partially-Observable Markov Decision Process (POMDP) and the requirements on positioning accuracy of all the UAVs are satisfied by use of the least collaborative information based on the accuracy threshold set by Cramer-Rao Lower Bound (CRLB).The simulation results show that compared with the non-screening situation the information screening strategy based on DRL can effectively enhance the real-time performance and operation efficiency of the collaborative positioning algorithm as well as reducing collaborative information redundancy while maintaining the positioning accuracy.