Abstract

Development of the Complex-Valued (CV) deep learning architectures has enabled us to exploit the amplitude and phase components of the CV Synthetic Aperture Radar (SAR) data. However, most of the available annotated SAR datasets provide only the amplitude information (Only detected SAR data) and disregard the phase information. The lack of high-quality and large-scale annotated CV-SAR datasets is a significant challenge for developing CV deep learning algorithms in remote sensing. In order to tackle this problem, a large-scale semantically annotated CV-SAR dataset is developed using the Single Look Complex (SLC) StripMap (SM) Sentinel-1 (S1) SAR data in two polarization channels (HH and HV) for Complex-Valued Deep Learning applications (S1SLC_CVDL). The S1SLC_CVDL dataset comprises 276,571 CV-SAR patches (100×100 pixel), derived from three scenes acquired over Chicago and Houston in the Uniate States, and Sao Paulo in Brazil in May 2021. These three scenes are selected to cover different landcovers including various vegetation covers, constructed areas and water bodies. The CV-SAR patches in this dataset are semantically annotated in 7 different classes, including, Agricultural fields (AG), Forest and Woodlands (FR), High Density Urban Areas (HD), High Rise Buildings (HR), Low Density Urban Areas (LD), Industrial Regions (IR), and Water Regions (WR). Refer to the cited articles for more information about the dataset and the selected S1 scenes.

Overall, the S1SLC_CVDL dataset provides semantically annotated CV-SAR data which can serve as a valuable resource for researchers and practitioners in the field of CV deep architecture developments for remote sensing applications.

Instructions:

The S1SLC_CVDL dataset comprises 276,571 patches (100×100 pixel) of Single Look Complex (SLC) StripMap (SM) Sentinel-1 (S1) CV-SAR data, derived from three scenes acquired over Chicago and Houston in the Uniate States, and Sao Paulo in Brazil in May 2021. Refer to the cited articles for more information about the dataset and the selected S1 scenes.

The S1SCL_CVDL.zip file, includes three subfolders (one for each S1 scene, Chicago, Houston, and Sao Paulo) containing the patches in two polarization channels (HH and HV) and the semantic labels of the corresponding patches. The data and label files are in .npy format and can be loaded into the python environment, using the “numpy.load(‘path to the file’)” function.

The semantic labels are provided as the numeric format as following:

Agricultural fields (AG)
Forest and Woodlands (FR)
High Density Urban Areas (HD)
High Rise Buildings (HR)
Low Density Urban Areas (LD)
Industrial Regions (IR)
Water Regions (WR)

For example, if the ith element of the label file is “1”, it indicates that the ith element in the corresponding data file for HH and HV polarization channels is from Agricultural fields (AG) class.

References

Please cite the following articles if you find the dataset useful:

R. M. Asiyabi, M. Datcu, A. Anghel, H. Nies, " Complex-Valued End-to-end Deep Network with Coherency Preservation for Complex-Valued SAR Data Reconstruction and Classification," in IEEE Transactions on Geoscience and Remote Sensing, 2023.
R. M. Asiyabi and M. Datcu, "Earth Observation Semantic Data Mining: Latent Dirichlet Allocation-Based Approach," in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 15, pp. 2607-2620, 2022, doi: 10.1109/JSTARS.2022.3159277.

Funding Agency:

European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie

Grant Number:

860370

Dataset Files

Dataset S1SLC_CVDL.rar (12.24 GB)

LOGIN TO ACCESS DATASET FILES
Open Access dataset files are accessible to all logged in users. Don't have a login? Create a free IEEE account. IEEE Membership is not required.