Research On Weak Target Detection Technology In Sea Clutter

With the wide application of high-resolution radars,the traditional target detection algorithms have obvious performance bottlenecks in the detection of weak targets on the sea surface.The echo of small slow target on the sea surface is weak,and the radar scattering area is too small to make the signal-to-clutter ratio very low,so the traditional adaptive detection algorithm is difficult to work.In complex space-time varying sea clutter environment,high resolution observation cannot meet the statistical characteristics of traditional large scale,and it is difficult to establish accurate target model;In the large amount of data obtained from sea observation,the categories of target and clutter are not balanced,and the sparsity of target relative to sea clutter makes sea clutter detection a challenge in traditional machine learning and pattern recognition.Therefore,weak target detection under the background of sea clutter has become the current research focus in the field of radar target detection.In this thesis,based on shore-to-sea warning radar,the research on the adaptive detection technology of weak targets in sea clutter is carried out,and the design of weak targets detection algorithm of sea clutter based on machine learning and deep learning algorithm is discussed.The simulation of IPIX sea clutter data verifies the actual performance of the algorithm,which provides a reference for engineering application.The specific content and research results are as follows:1.The problem of adaptive detection of weak targets under the background of sea clutter is studied.The characteristics analysis of sea clutter distribution,statistical modeling,parameter estimation,clutter model matching,detector design and detector selection are studied,and the whole process from sea clutter modeling to adaptive target detection is realized.The process goes through.Explored the IG-CG distribution under the compound Gaussian model,analyzed the different characteristics of the IG-CG distribution under single parameter and dual parameter,and the difference between the distribution models under the two different speckle components of IC-CG,and realized different scenarios.The construction of the model library of the sea clutter.Deduced the detectors under a variety of compound Gaussian distributions under the framework of AMF,AMNF and generalized likelihood ratio,focusing on the compound Gaussian model,the texture components are Gamma distribution and inverse Gamma distribution,and the speckle components are complex Gaussian distribution and Rayleigh The generalized likelihood ratio adaptive detector algorithm under the two distribution models of IG-CG distribution,K distribution and GP distribution,experiments have verified that the performance of the GLRT detector whose speckle component is a complex Gaussian distribution is better than that of speckle.The GLRT detector is distributed well,and it has better performance than the traditional CA-CFAR detector.2.The weak target detection technology of sea clutter based on machine learning and deep learning is studied.Based on the signal feature analysis of IPIX measured sea clutter data,a machine learning feature vector space was constructed with three features of information entropy,frequency domain peak to average ratio and Hearst index.A controllable false alarm rate detector based on the XGBOOST algorithm is designed.The performance of the detector under different clutter areas and different signal-to-clutter ratios is simulated and analyzed,and it is compared with traditional support vector machines,random forests and other machine learning algorithms.Experimental comparison shows that the algorithm based on XGBOOST has a better detection rate.At the same time,the time-frequency characteristics of sea clutter data are analyzed,and the time-frequency characteristics of sea clutter are used as the training set.A deep learning detector based on the LeNet network model is designed.By comparing with machine learning and adaptive detector,the simulation proves that the LeNet network model has better performance in heavy clutter region.