Island's Remote Sensing Information Extraction Technology Of Polarimetric SAR Based On Deep Learning

The island scattered in the ocean is not only rich in natural resources,but also an important support for China's implementation of general secretary Xi Jinping's strategic thinking of maritime power,the development of marine economy,the protection of maritime rights and interests and the maintenance of coastal defense safety.Because of the advantages of large amount of information,fast information acquisition,all day,allweather observation and so on,synthetic aperture radar(SAR)can well cope with and meet the complex surface environment and fine regulatory requirements of islands,and has become an important means of island investigation.Since the 21 st century,the advantages of deep learning in remote sensing information extraction have become increasingly prominent,which has become a hot issue for scholars all over the world.Compared with the traditional machine learning method,deep learning can automatically mine more abundant feature information,and can more comprehensively describe the object,so as to improve the level of island remote sensing information extraction.In recent years,some scholars have applied deep learning methods to the extraction of surface feature information in the synthetic aperture radar island.However,the types of features in the island are complex,artificial features and natural features coexist,and their texture,shape,and distribution characteristics are different.At present,there is still no method that can accurately extract all types of island's features.Therefore,this article takes "Deep Learning-based Polarization SAR Island Remote Sensing Information Extraction Technology" as the topic,and does the following work:1.According to the features of the island,the most suitable polarization decomposition method and filtering method for fully polarized SAR data are explored.2.An improved Alexnet deep learning network model is proposed to classify all polarimetric SAR Island features.On the basis of Alexnet,the convolution kernel size and full connection layer are adjusted to reduce the parameters;In order to adapt to the characteristics of SAR data and island features better and improve the accuracy and adaptability to different data,pooling layer is added to reduce the dimension and computational complexity.Compared with the classification results of AlexNet,the accuracy is improved by 3.75%.3.The 152 layer deep residual network model is improved,and the improved deep learning network model is applied to polarimetric SAR island remote sensing information extraction.Residual blocks are used to solve the problem of gradient disappearance and degradation caused by too deep network depth.The essential characteristics of data are deeply mined to improve the performance and robustness of network model.The classification accuracy is improved by 7.42%.The results show that the deep learning method has obvious advantages in the field of remote sensing feature information extraction,and the improved Alexnet can effectively distinguish multiple types of island features in the full polarimetric SAR data,which is better than the classical Alexnet.The deep residual network can distinguish the island features well,and its accuracy is better than that of Alexnet.It has great research value for island remote sensing information extraction.