The advancement in unmanned aerial vehicle (UAV) technology has increasingly drawn the attention of researchers to cooperative operations among multi-UAVs, with path planning standing out as a critical challenge. This paper proposes a method leveraging the multi-agent soft actor-critic algorithm enhanced by squeeze-and-excitation attention and prioritized experience replay (MASAC-SEPR) to address the heterogeneous multi-UAVs path planning problem. The model is formulated as a partially observable Markov decision process (POMDP), integrating an advanced algorithmic framework that incorporates the Squeeze-and-Excitation Attention (SEAttention) module, actor and critic networks. Additionally, it employs a prioritized experience replay memory mechanism to ensure efficient training. Simulation results demonstrate that this model surpasses traditional approaches in terms of performance, convergence speed, and adaptability to complex environments. It facilitates efficient and accurate cooperative path planning for heterogeneous multi-UAVs, providing robust algorithmic support for practical applications.