부산대학교 도서관

In this paper, we present the SARFish challenge and dataset. The challenge focuses on the use of Synthetic Aperture Radar (SAR) imagery for the identification of vessels involved in illegal, unreported and unregulated (IUU) fishing which damages ecological systems and causes losses for fishing industries and governments worldwide. The SARFish dataset is a free and open large-scale complex-valued SAR dataset which is based upon Sentinel-1 imagery and built upon the xView3 labels. We expect this dataset to help in advancing the state of the art in automated ship de-tection from SAR imagery, contextual representation learning, and the application of deep complex-valued neural networks. We also hope the availability of the SARFish dataset will stimulate developments on other topics of interest that can naturally tackle complex-valued data such as quantum-inspired approaches.