AIR-PolSAR-Seg-2.0: Polarimetric SAR Ground Terrain Classification Dataset for Large-scale Complex Scenes

The AIR-PolSAR-Seg-2.0 dataset aims to build a polarimetric SAR ground terrain classification dataset for large-scale complex scenes, promoting research and development in polarimetric synthetic aperture radar ground terrain classification within the field of SAR image interpretation. The dataset is composed of GF-3 L1A complex SAR images from three different regions, with a spatial resolution of 8 m and includes four polarimetric modes (HH, HV, VH, and VV), covering six typical ground terrain categories: water, vegetation, bareland, buildings, roads, and mountains. The dataset exhibits characteristics of complex scenes, large scale, diverse scattering intensities, irregular boundary distributions, varying category scales, and unbalanced sample distribution. Additionally, the complete SAR images of the three scenes have been cropped into 24,672 slices of 512×512 pixels to facilitate experimental verification within a deep learning framework.

AIR-PolSAR-Seg-2.0数据集旨在构建面向大规模复杂场景的极化合成孔径雷达（Polarimetric SAR）地物分类数据集，以推动SAR图像解译领域中极化合成孔径雷达地物分类相关研究与技术发展。该数据集由来自三个不同区域的GF-3 L1A级复SAR图像构成，空间分辨率为8米，包含HH、HV、VH、VV四种极化模式，涵盖水域、植被、裸地、建筑物、道路、山地六类典型地物。该数据集具备场景复杂、规模庞大、散射强度多样、边界分布不规则、类别尺度各异以及样本分布不均衡等特点。此外，为便于在深度学习框架中开展实验验证，三幅完整的SAR图像已被裁剪为24672张512×512像素的图像切片。

搜集汇总

背景与挑战

背景概述

AIR-PolSAR-Seg-2.0是一个针对大规模复杂场景的极化SAR地面地形分类数据集，包含GF-3 L1A复杂SAR图像，覆盖六种典型地形类别，具有8米空间分辨率和四种极化模式。数据集已裁剪为24,672张512×512像素的切片，适用于深度学习研究。