<p>Phase unwrapping is a critical technique in interferometric synthetic aperture radar(InSAR) applications, with deep learning being widely adopted to address this problem. Compared to traditional algorithms that rely on the assumption of phase continuity, deep learning methods offer more efficient phase unwrapping. However, existing approaches still face challenges including inadequate performance in complex terrains, limited model generalization, and failures in low-coherence regions. To address these issues in deep learning phase unwrapping, we designed RFAUNet (Receptive Field and Attention U-Net), a robust network integrating receptive field modules and attention mechanisms. Built upon the U-Net architecture, RFAUNet incorporates the Convolutional Block Attention Module (CBAM), Atrous Spatial Pyramid Pooling (ASPP), and the Receptive Field Block (RFB). RFAUNet is trained using two types of simulated datasets and optimized by minimizing the root mean square error between the true phase and the unwrapped phase, facilitating direct phase unwrapping. Experimental results demonstrate that RFAUNet exhibits superior accuracy in challenging low-coherence areas and complex terrains, while maintaining high phase unwrapping efficiency, and real-time potential on both simulated and real-world InSAR data.</p>