Research On Sea Ice Classification Method Of Polarimetric SAR Based On Deep Learning

With recent reductions in Arctic ice extent,there has been growing economic interest in shipping and natural resource extraction in the Arctic.To support safe Arctic operations and navigation in ice-infested waters,timely high-resolution information of the ice coverage is crucial.Synthetic Aperture Radar(SAR)can generate observations regardless of the weather conditions(cloud,rain)and sunlight,thus can provide abundant target scattering information for sea ice classification,which has great potential in sea ice classification.At the same time,deep learning method has the ability to automatically learn features from images,and has become an indispensable research method in image processing field.In this paper,sea ice detection method of polarimetric SAR based on deep learning was explored.The main research works are summarized as follows:1.The establishment of typical sea ice sample library from polarized SAR images.Deep learning based sea ice classification can only be facilitated by the establishment of sea ice sample library.At present,sea ice samples are mainly identified by visual method,which is time-consuming and low precision.This paper summarizes the sea ice classification standards and ice interpretation methods which are commonly used at home and abroad,and a sea ice sample library is established according to the characteristics of Polarimetric SAR sea ice images.2.A Complex Valued Convolutional Neural Network(CV-CNN)based classification method from polarized-SAR images was proposed.Convolutional Neural Network(CNN)has great potential in image classification and recognition.In order to make full use of the unique phase information in Polarimar SAR images,the complex number operation was introduced into the CNN model,and a CV-CNN based Polarimar SAR sea ice image ice-water classification method was proposed.Experimental results show that the proposed method can significantly improve the accuracy of ice and water classification in Polarimetric SAR sea ice images.3.A multi-class sea ice classification method based on the optimized Deeplab V3+network was developed for polarized SAR images.Aiming at the problem of fuzzy classification boundary in semantic segmentation network,a multi-class sea ice classification method based on optimized Deeplab V3+ network was proposed by combining superpixel segmentation with classical semantic segmentation model Deeplab V3+ network for Polar SAR sea ice classification.The experimental results show that the optimized Deeplab V3+ network has great application potential in polarization SAR sea ice image classification.