The transformer-based methods have demonstrated remarkable advancements in synthetic aperture radar (SAR) classification. Nevertheless, many of these methods ignore global statistical information and semantic feature interaction for effectively characterizing different SAR land covers under complex structures. Leveraging second-order statistics presents an efficacious approach to well characterize the statistical features of SAR patches. Motivated by this, we integrate pyramid pooling and global covariance pooling techniques into each of the multihead self-attention blocks, thereby facilitating the extraction of powerful contextual features and the global statistical nature of SAR patches, namely multiorder pooling transformer module. Simultaneously, a semantic feature aggregation module is utilized for capturing local deep features and modeling the interaction of feature information across various feature levels. Both of these modules are embedded into a U-shaped architecture, which we refer to as a multiorder pooling transformer-based semantic feature aggregation network (MPT-SFANet). Extensive experimental results on TerraSAR, Sentinel-1B, and GF-3 SAR image classification datasets indicate that MPT-SFANet exceeds several relevant methods.

中文翻译：

---

MPT-SFANet：用于 SAR 图像分类的基于多阶池化变压器的语义特征聚合网络


基于变压器的方法在合成孔径雷达（SAR）分类方面取得了显着的进步。然而，这些方法中的许多方法忽略了全球统计信息和语义特征交互，以有效地表征复杂结构下不同的SAR土地覆盖。利用二阶统计数据提供了一种有效的方法来很好地表征 SAR 补丁的统计特征。受此启发，我们将金字塔池化和全局协方差池化技术集成到每个多头自注意力块中，从而有助于提取强大的上下文特征和SAR补丁的全局统计性质，即多阶池化变压器模块。同时，利用语义特征聚合模块来捕获局部深层特征并对跨不同特征级别的特征信息的交互进行建模。这两个模块都嵌入到 U 形架构中，我们将其称为基于多阶池化变压器的语义特征聚合网络（MPT-SFANet）。 TerraSAR、Sentinel-1B 和 GF-3 SAR 图像分类数据集上的大量实验结果表明 MPT-SFANet 超越了多种相关方法。