Polsarhub: A Large-Scale Repository for Polsar Data

In recent times, machine learning algorithms are widely utilized by researchers for the classification of PolSAR data. The availability of PolSAR data from sources such as UAVSAR, Sentinel, and now the NISAR mission has expanded the potential applications of PolSAR data. Efficient handling of this extensive data is achieved through deep learning models, leading to an increase in demand for a substantial amount of training data. The collection of labelled data poses a significant challenge, as obtaining ground truth for every geographical area is difficult. While repositories like ImageNet offer labeled datasets for optical data, they are unsuitable for PolSAR data. Therefore, this paper proposes the “PolSARHub” repository as a proposed solution specifically designed for PolSAR data. PolSARHub will provide the labelled data for all land-cover features like waterbodies, settlement, forest, mangroves, saltpans, and grasslands etc. It will take care of different sensors with varying wavelengths, incidence angles, resolution, polarization, orientation angle, and dielectric properties. The labelled data will be generated from time series data to achieve the said goals.