Extraction Of Coastal Aquaculture Ponds From Remote Sensing Images Based On U~2-Net Deep Learning Model

As the main production mode of aquaculture,coastal aquaculture ponds are of great significance in promoting economic development and ensuring food supply and their rapid development has also led to problems such as damage to natural resources and the ecological environment.Rapid and accurate acquisition of spatiotemporal distribution information of coastal aquaculture ponds can provide support for the scientific utilization of marine resources and the sustainable development of the ecological environment.At present,many studies carried out information extraction models on coastal aquaculture ponds using remote sensing technology,but the extraction results are still affected by the effect of ’same spectral foreign objects’.In order to solve the above problems,this dissertation proposed a remote sensing information extraction model of coastal ponds based on the U2-Net deep learning network,completed the remote sensing information extraction of coastal aquaculture ponds in the Zhoushan archipelago from 1984 to 2022,analyzed the spatiotemporal evolution of coastal aquaculture ponds in Zhoushan archipelago and developed the remote sensing information extraction system for coastal aquaculture ponds.The main conclusions are as follows:(1)This dissertation proposed a remote sensing information extraction model of coastal ponds based on the U2-Net deep learning neural network.The model used spectral feature analysis to construct stacking rules,which reduced the misdivision rate of coastal aquaculture ponds and other types of water bodies.The results showed that the accuracy rate was 93.23%,the recall rate was 94.76%,and the F-measure was 0.94.(2)From 1984 to 2022,the area of aquaculture ponds along the coast of the Zhoushan archipelago showed an overall expansion trend.After almost 40 years of change,from 471.21 ha in 1984 to 3668.55 ha in 2022.The first stage was from 1984 to 2004,and the area of aquaculture ponds along the coast of the Zhoushan archipelago increased;The second stage,from 2005 to 2022,showed a stable and slightly decreasing trend in the area of aquaculture ponds.The islands of the Zhoushan archipelago mainly showed three different types of change trends:1)Continuous increase.Such as Daishan Island,Liuheng Island,Sijiao Island,Zhujiajian Island,Taohua Island,Changtushan Island,and Qushan Island.2)Gradually decrease.Such as Jintang Island and Xiushan Island.3)Increase first and then decrease.Such as Zhoushan Island.(3)From 1984 to 2022,the culture process of aquaculture ponds along the coast of the Zhoushan archipelago was remarkable.Two-year coastal aquaculture ponds are the most distributed in the archipelago,accounting for about 18.51%of the total coastal pond area.Less than 15%of coastal ponds have a process of more than 30 years.Coastal ponds of 53.55 ha were observed throughout the study period.In this dissertation,we constructed a remote sensing information extraction model of coastal ponds based on the U2-Net deep learning network,which is of great significance for better-formulating aquaculture development planning and promoting the sustainable development of aquaculture and coastal areas.