Arctic Sea Ice Classification Based On Sentinel-1 Polarization Data

Information on sea ice in polar region is crucial for a variety of applications,including climate change research and navigation.As an integral part of the Earth's climate system,sea ice interacts with the ocean and atmosphere,regulating the flux of heat and water.Arctic sea ice helps to moderate the global climate by maintaining freezing temperatures in the polar region.But the trend of Arctic sea ice in winter and summer has been negative in recent decades.As an emerging learning method,deep learning learns the features of images through training,and has attracted the attention of all walks of life for its high-precision classification effect.The method based on sea ice deep learning classification can automatically at the same time in the training of effective characteristics of sea ice is extracted,greatly reduces the workload of artificial process data,effectively avoids the complex characteristics of the project,can be classified according to the characteristics of arctic sea ice,enhance the accuracy in classification.The main research contents of this paper mainly include the following three points:(1)As the research of sea ice classification based on deep learning does not attempt to conduct sea ice classification research on a large range of images,this paper designs CNN and FCN sea ice classification methods that are suitable for a large range of Arctic sea ice classification based on residual learning.Explore how to use deep learning method to process a large number of Sentinel-1 sea ice data quickly and efficiently,so as to obtain effective sea ice information and achieve the task of sea ice image classification.(2)In view of the limited sea ice classification algorithms developed based on Sentinel-1data in recent years and the lack of research on sea ice classification with different polarization modes based on deep learning algorithm,this paper uses FCN sea ice classification method to test the sea ice classification performance of Sentinel-1 data under three different polarization modes.We compare the effects of different polarization modes on sea ice classification and analyze the ability of different polarization modes to identify different types of sea ice.(3)In order to solve the problems related to the lack of research on the migration learning of large range of sea ice using deep learning method,in this paper,the Sentinel-1 HH+HV polarization data are used to study the migration of summer and winter sea ice in the Arctic Ocean using the FCN sea ice classification method.By comparing the results of winter and summer sea ice classification,the migration time and migration ability of the FCN sea ice classification method are evaluated.The time transfer learning ability of the deep learning-based Arctic sea ice classification method are verified.In this paper,Sentinel-1 polarization data is used as the data source to classify the sea ice in the Arctic Ocean.The aim is to use the deep learning algorithm which has excellent performance in the field of image classification,and use the Sentinel-1 polarization data in the Arctic Ocean to classify the sea ice in a large range of the Arctic Ocean according to the characteristics of different types of sea ice in the Arctic Ocean.We verify the classification performance of the sea ice classification method based on deep learning,quickly and accurately produce sea ice maps,analyze the sea ice conditions of the Arctic Ocean sea ice,and provide fast and accurate sea ice maps for climate change research and Arctic ship navigation.