Hierarchical pyramid refined U-Net for creating the first 2 m resolution multi-class national-scale spatial distribution dataset of offshore observable marine aquaculture in China

Precise extraction and dynamic monitoring of observable offshore marine aquaculture (OOMA) areas in remote sensing imagery are essential for the rational layout of offshore aquaculture zones and assessing the marine ecological environment. At present, national-scale offshore marine aquaculture spatial distribution datasets derived from remote sensing imagery in China are of medium to low resolution, and there remains a lack of high-spatial-resolution (HSR) dataset. To address this gap, this study proposes a novel semantic segmentation framework, Hierarchical Pyramid Refined U-Net (HPR-UNet), based on multi-source 2 m resolution remote sensing imagery for the multi-scale extraction of OOMA. The first HSR OOMA spatial distribution dataset of China (RCdata_2022_2 m) has been generated. The study reveals the following: (1) The proposed method is effective in extracting multi-scale OOMA, especially small-scale ones. (2) The RCdata_2022_2 m achieved an overall accuracy of 97.27%. Compared to medium- and low-resolution datasets, RCdata_2022_2 m demonstrated higher accuracy and better extraction results. (3) From 2022 to 2023, the total area of OOMA in China was approximately 253,096.08 ha, with raft aquaculture occupying 232,523.69 ha, and cage aquaculture covering 20,572.39 ha.