Abstract: Compared with manned aircraft, unmanned aerial vehicles (UAVs) are affordable and convenient for high-risk missions. Therefore, UAVs are increasingly playing an important role in military and civilian fields. Today, UAVs have been exploited to perform special missions carrying some important equipment. However, influenced by the constraints of single UAV's performance and load, it has become a research hotspot that multi-UAVs perform search cooperatively. The process is to minimize the uncertainty of the unknown area and to find the target as much as possible. In terms of cooperation among UAVs, the more effective method based on search map is used. And search process optimization on the basis of distributed model predictive control (DMPC) or traditional swarm intelligence algorithms are adopted, but these methods have some limitations. Due to the behavior of swarm intelligent individual have the characteristics of the decentralization, distribution, and overall self-organization, which match with the requirements of the localization, distribution and robustness of the UAV cooperate search. Nevertheless, the traditional swarm intelligence algorithms have low search efficiency and are easy to fall into local optimum. To solve the problem of repeated search, static targets and low efficiency in cooperative search for multi-UAVs, a method based on improved pigeon-inspired optimization (PIO) and Markov chain was proposed. Firstly, a honeycomb environmental model similar to the sensor detect region was established to reduce repeated search for the area. Secondly, Markov chain with the Gaussian distribution was used to represent dynamic movement of targets. Thirdly, Cauchy mutation and Gaussian mutation were introduced into the map and compass operator and the landmark operator of PIO, respectively. Meanwhile, simulated annealing (SA) mechanism was exploited to reserve the worse individual, which effectively reduced the problem that PIO was easy to fall into local optimum. Finally, the algorithm was compared with other swarm intelligence algorithms through simulation experiments. The results show that the new method is effective and available.