SAR Image Classification Based On Deep Learning And Sparse Representation

Synthetic Aperture Radar(SAR)is an active coherent imaging radar,capable of observing and imaging a large area of the whole earth all day,with strong penetrating power and high resolution.For remote sensing information is an extremely important means,in the military field and civil neighborhood have important application value.SAR in the military field of the most important application is to achieve the detection and identification of specific military targets,with the ever-increasing information collection capabilities,by means of machine learning or pattern recognition and other cutting-edge technology for synthetic aperture radar images automatically or semi-automatic solution Translation can greatly improve the efficiency of data processing.With the development of radar imaging technology,we can get a large number of remote sensing images with high resolution,so the target detection and recognition for SAR images has become a hot research topic both at home and abroad.This thesis focuses on the application of deep learning and sparse representation of the model and the SAR image classification task.SAR image classification problem,refers to the machine through a variety of learning algorithms to automatically manage the various types of images for effective management and classification.As a new machine learning method proposed in recent years,deep learning has been widely focused on its rapid development and has been widely used in many fields such as speech recognition,image recognition and natural language processing.The depth learning model has the ability of hierarchical learning.Compared with some shallow feature learning models,the depth learning model learns the deep characteristic of the original data through the multi-layer nonlinear network structure,which is the original data Abstract expressions,which effectively characterize the inherent information of the original data.The sparse representation model is designed to find an efficient expression for the data.It is pointed out that the human visual system has the characteristics of sparse representation of the image,and the image classification based on the sparse representation method has become one of the important research hotspots in the field of pattern recognition The Wright et al.First applied the sparse representation to face recognition tasks and achieved better results than traditional methods.Since then,the use of sparse representation of the image classification work has been more and more attention.The main work of this thesis is as follows:The Pca Net model and its application in image classification are introduced to introduce the pooling operation in the convolution neural network,which reduces the robustness of the network while reducing the data calculation pressure and accelerating the classification efficiency.On the basis of the first work,the SAR-SIFT feature description operator is introduced for the SAR image's own characteristics,which makes the improved network more effective when dealing with SAR images.This thesis studies the excellent performance of the convolution sparse coding model in high resolution image reconstruction,and introduces it into the work of image classification.A SAR image classification method based on convolution sparse coding is proposed.The difference between the method and the general sparse representation method in image classification is analyzed experimentally.