Datasets
Standard Dataset
high-resolution optical and SAR for image retrieval

- Citation Author(s):
- Submitted by:
- Lingyun Gu
- Last updated:
- Fri, 03/21/2025 - 11:36
- DOI:
- 10.21227/7sjk-ff76
- License:
- Categories:
- Keywords:
Abstract
The HROS dataset consists of OPT and SAR images collected from multiple sources. The OPT images are collected from the large-scale public remote sensing datasets. Image sources include different platforms, including Google Maps, JL-1 satellite, Gaofen-2 satellite, aerial imagery, etc. The SAR images are captured by the AS-01 satellite, developed by Skysight Technology Co., LTD. The AS-01 satellite is equipped with a two-dimensional scanning plane solid-state active phased array synthetic aperture radar (SAR) payload, which enables high-resolution imaging capabilities. The allocated SAR images involve three different SAR imaging modes, include sliding cluster SAR imaging, strip SAR imaging and scan imaging. There are three different spatial resolutions for SAR images: 0.5 m, 1 m, and 3 m, enabling a comprehensive study of fine-grained cross-modal retrieval. The dataset covers a wide geographic range, primarily focusing on harbor areas and surrounding urban environments. The SAR data were acquired between August 2023 and December 2024, ensuring a temporally diverse dataset.
The dataset contains two folders, optical images and SAR images. The structure of each dataset is the standard classification dataset structure: root- train/test- class-xx.jpg.