<p>Ship classification in Synthetic Aperture Radar (SAR) images remains challenging due to the imaging characteristics of SAR and subtle differences among certain ship categories. Although polarization information in SAR data offers additional discriminative cues, it has not been fully exploited in prior studies. To address this, we propose an adaptive fusion network with polarization-guided hybrid attention (PGAF-Net) for dual-polarized SAR ship classification. PGAF-Net comprises two key modules: the dual-polarization feature extraction (DPFE) module and the multi-frequency adaptive fusion (MFAF) module. The DPFE module incorporates polarization-data attention (PDA) blocks, where the polarization-driven branch utilizes high-level semantics of polarization features to activate local semantic regions of ship targets and reweights features accordingly. The MFAF module employs multi-frequency attention (MFA) to adaptively fuse dual-polarized features, enabling the preservation of discriminative information across different frequency components. In addition, a joint loss function that combines the cross-entropy loss and the island loss is employed to enhance inter-class separability and reduce intra-class variations in the feature space. Experimental results on the OpenSARShip2.0 dataset indicate that PGAF-Net achieves competitive performance and outperforms several state-of-the-art methods for dual-polarized SAR ship classification.</p>