Cooperative planning of multiple unmanned aerial vehicles (UAVs) swarm is a hotspot research issue in the field of UAVs autonomous control. A novel task allocation method is proposed for multi-UAV reconnaissance based on unsupervised learning discrete pigeon-inspired optimization. Firstly, a model of UAV is established by Dubins path. Secondly, a simplified sensor model and a reconnaissance target model are given, and a task allocation model of multi-UAV and index function of performance are established. The reconnaissance targets are grouped flexibly by the unsupervised learning method, and then the task allocation model is solved by using the improved discrete pigeon-inspired optimization strategy, effectively solving the unbalanced task load between UAV and improving the efficiency of UAV swarm reconnaissance. Finally, the method proposed in this paper is valid through the simulation comparison experiments and the three dimensional situational scene simulation platform experiments.