Research And Applications Of SAR Imaging For Sea Surface Feature Extraction Based On Compressed Sensing

The ocean,which occupies most of the earth’s area,is playing an increasingly important role in national production and people’s lives.Observing and deeply understanding the information on the sea surface is of great scientific significance for the sustainable development of human production and life.Synthetic Aperture Radar(SAR)provides an effective tool for sea surface research with its all-time,all-weather and high-resolution remote sensing characteristics.Through the analysis of remote sensing data,ocean features such as eddies and oil spill,especially some small and meso-micro scale features,can be extracted.These features are important research objects for us to obtain ocean information.But SAR data requires huge storage resource and processing costs.In addition,the features of ocean have complex shape.So a large amount of data must be used for modeling and estimation,which limits the feasibility of analyzing ocean features through SAR images.This paper focuses on how to reduce the amount of SAR sea surface imaging data,achieve high-resolution image under high compression ratio,and obtain more reliable sea surface information faster,according to the application of ocean feature extraction,and combined with the characteristics of SAR data,studies the sparse representation method based on sparse equalization and the construction method of low column correlation measurement matrix,and uses it for sea surface compressed sensing imaging.On this basis,the Microcanonical Multifractal Formalism is used to directly obtain ocean features from images,which provides technology for efficient monitoring of sea surface features caused by nature or humanity such as ocean eddy and oil spills.In response to the above problems,the main work of this article is as follows:(1)By studying signal sparsity and compressibility,the condition and the process of sparse representation are analyzed.The sparse representation methods of two-dimensional data are studied.Furthermore,the idea of sparse equalization strategy is analyzed.Then a sparse representation method is proposed by this idea.The measure dimension is reduced by sparse averaging between columns,and the reconstruction efficiency and performance are improved.The simulation results show that this method has universal applicability in compressed sensing of image and Ocean echo datao of SAR.It can significantly improve the reconstruction effect under low measure dimension,shorten the reconstruction time,and improve the performance of compressed sensing.So it can provide an effective method for low-dimensional measurement of big data.(2)With the study of the zero-space characteristics,Restricted Isometry Property,correlation of the measurement matrix and their relationship,we conclude in theoretical that the low column correlation measurement matrix can meet the Restricted Isometry Property with higher probability,and reconstruct the original signal through fewer iterations.A method for constructing a low mutual coherence matrix is proposed.By maintaining the correlation between the columns of the entire measurement matrix at a low level,signal can be compressed at low sampling rate and reconstructed under fast iteration.The reconstruction effect is improved without increasing the running time.This effective measurement matrix construction method is provided for the real time and low sampling rate compressed sensing of massive data.(3)Arming at feature extraction,the sea surface SAR imaging technology based on compressed sensing is studied,and then the sea surface Normalized Radar Cross Section(NRCS)imaging based on compressed sensing is realized.This method is improved by sparse equalization technology and the low column correlation measurement matrix.The simulation results verify that this method can generate high-resolution sea surface image at a higher compression ratio and preserve as much information as possible for ocean feature analysis.(4)The multifractal structure in the sea surface NRCS image is studied.Considering the characteristics of the sea surface image and the requirements of the ocean feature analysis,the singularity analysis method is optimized and Microcanonical Multifractal Formalism is applied.In order to further obtain the main features of the ocean,Most Singular Manifold method is studied and used,and the most singular threshold is determined.The simulation results show that the distribution of sea surface information such as manifolds and eddies can be obtained with this method,and the Most Singular Manifold method with appropriate thresholds can extract the main features of ocean.This method is also employed in remote sensing image.The results prove the universality of it.By this way,rapid oil spill detection can also be carried out based on the reflection mechanism of oil films.