Research On Radar Target Recognition Based On Manifold Learning

Synthetic Aperture Radar(SAR)is a kind of imaging radar which has a series of advantages,such as high precision,high resolution,all-weather and strong penetration ability.With the rapid development of technology,SAR image contains more and more information,high-dimensional data processing has become a major focus of image processing.Data dimensionality reduction technology is the key technology in the process of feature extraction of image data.Because every data of SAR image is a high-dimensional data set,this situation usually leads to the problem of "Dimension Disaster".In order to solve the dimensionality problem,it is necessary to reduce the dimensionality of high-dimensional data sets when processing the original data,which is a very important step.As a machine learning method,manifold learning is widely used in data visualization,artificial intelligence,face recognition and image retrieval.At the same time,it is also a dimension reduction method.Based on the characteristics of SAR image and the research status of manifold learning method,this paper starts a series of work from the application and improvement of algorithm.The main research work and research contents are summarized as follows:(1)Combined with the traditional data dimensionality reduction method,two classical local manifold learning methods are studied and improved.The implementation process,derivation principle and time complexity of the algorithm are analyzed and compared.(2)Based on the Laplacian mapping method,the concept of stochastic process is introduced,and the Local Linear Embedding method and the Stochastic Laplacian mapping method are fused.The algorithm is applied to MSTAR dataset and classified by KNN classifier.The experimental results show the effectiveness of the local manifold learning method.(3)On the basis of deeply studying the traditional single manifold learning method,this paper introduces the multi-manifold learning method,assuming that the data sets of different categories in the high-dimensional space exist on different manifold structures,and embeds different low-dimensional manifolds into the high-dimensional data sets of different categories,and improves the classification label of the multi-manifold LE method,applies the multi-manifold learning method and constructs a classifier for classification,experimental results verify the effectiveness of the algorithm.