Mapping the distribution and dynamics of coastal aquaculture ponds using Landsat time series data based on U2-Net deep learning model

Abstract

ABSTRACTThe extraction of coastal aquaculture ponds from remote sensing images is affected by the effect of ‘same object with different spectrums’ and ‘different objects with same spectrums’. The U2-Net deep learning network is able to capture more contextual information from different scales to solve this problem and is suitable for salient object detection. This study proposed a remote sensing information extraction model of coastal ponds based on U2-Net deep learning network, completed the remote sensing information extraction of coastal aquaculture ponds in Zhoushan Archipelago from 1984 to 2022, analyzed the spatiotemporal evolution of coastal aquaculture ponds in Zhoushan Archipelago. The results showed that the developed model was more accurate with a precision rate of 96.12%, a recall rate of 95.43%, and an F1-measure of 0.96. During the study period, the area of the aquaculture ponds in the Zhoushan Archipelago demonstrated an increasing trend, expanding from 471.21 hm2 in 1984–3668.55 hm2 in 2022. In addition, over half of the aquaculture ponds had been active for more than 10 years. The method developed in this study is capable of rapidly and accurately mapping coastal aquaculture ponds, and thus is significant for the management of marine resources and promoting sustainable development.

Keywords

• Coastal aquaculture ponds
• U2-Net
• deep learning model
• spatiotemporal patterns
• accuracy assessment