In emergency disaster relief scenarios with inadequate communication infrastructure, traditional monitoring methods struggle to provide dynamic surveillance, necessitating efficient resource scheduling for UAV-assisted Mobile Edge Computing. This paper presents EAS-LRPO, a lightweight multi-UAV resource scheduling algorithm that addresses computational overhead and coordination challenges. By replacing the value network with an advantage function and introducing an attention-sharing mechanism (EAS), the method enhances policy sensitivity to critical states while reducing inference costs. The Link Relative Policy Optimization (LRPO) incorporates distance-weighted advantages and Sinkhorn-Wasserstein regularization to improve UAV swarm coordination and prevent overfitting. Simulation results demonstrate that EAS-LRPO achieves faster and more stable convergence in training/test loss compared to LRPO, DDPG, and PPO-CO. It also exhibits superior average resource utilization across varying task loads, highlighting its robustness in dynamic emergency environments.