Target Recongnition Of Ice Sheet In Ice Radar Image Based On Adversarial Network And Capsule Network

Polar ice sheet,including Antarctic ice sheet and Greenland ice sheet,accounts for99% of the total volume of glaciers in the world.The melting of ice and snow on polar ice sheets caused by global warming has led to sea level rise,and many coastal cities are at risk of flooding.Therefore,it is of great significance to study the change trend of polar ice sheet for global climate research.The research on the change trend of polar ice sheet needs to have a full understanding of the ice sheet interior.The relevant data of the internal target research of the ice sheet mainly rely on the ice radar technology,and the research of ice sheet underground targets mainly depends on the use of special radar detection tools to obtain a large number of radar image data and analyze them.The key step is to identify the internal targets of ice sheet.The main work of this paper is to solve the problem of limited by recognition accuracy,manual intervention and unable to achieve automatic classification in the current ice sheet internal target recognition algorithm.Combined with the generative adversarial network and capsule network,three ice radar image ice sheet internal target recognition methods based on deep learning are proposed.1.Ice radar image classification method of ice sheet internal target based on generative adversarial network.In this method,U-Net network structure is used as the adversarial network to generate the network,which effectively increases the expression of identifying network In this paper,the discrimination network for generating adversarial network is improved,and Patch network is used as the discrimination network to improve the network's ability to extract the detail information of ice radar image and enhance the discrimination ability.2.Ice radar image classification method of ice sheet internal target based on capsule network.In this method,the capsule network structure is used instead of convolution network to extract the feature information of the image,which can extract the spatial position information of the image more accurately and it is able to make full use of the regularity of the spatial distribution of the targets in the ice sheet in the ice radar image to improve the classification accuracy.3.Ice radar image classification method based on the SegCaps network.In this method,the filtering process is added to the SegCaps network which is used to classify the internal target of ice sheet ice radar image.Because the speckle noise has a great interference on the image,we use the filtering method to process the radar image,reduce the speckle interference and save the details of the ice radar image as much as possible.Finally,we propose an automatic classification of Ice sheet subsurface targets in radar sounder data,which realizes the automatic and accurate classification of the underground target in the ice layer.