Based On Sparse Representation For SAR Image Classification Model And Its Algorithm

Synthetic Aperture Radar(SAR)is widely used in the military and civil fields because of its penetrability,all-weather and high-resolution images.It is not affected by diurnal changes and climate change.However,the particularity of the SAR imaging mechanism leads to the inherent multiplicative speckle noise,which seriously affects the interpretation of SAR images.SAR image classification is one of the most important research contents in SAR image interpretation.The classification accuracy directly reflects the quality of SAR image interpretation.Therefore,the research of SAR image classification algorithm with high precision,high robustness and low time-consuming has been the focus of SAR image research.Aiming at solving the problems of the SAR image classification algorithm,such as low classification accuracy,high time complexity and low robustness,this paper constructs new classification model via combining the multiscale technology,superpixel and sparse representation.The meaningful research works include:1)A multiscale classification method with fusion feature strategy method is proposed for SAR image.Firstly,Sparse Representation(SR)method is used to describe the sparse characteristics of SAR image.In order to reduce the influence of noise on experimental results in the process of dictionary construction,the feature vector method is selected to construct a dictionary.Secondly,due to the complexity and diversity of SAR image features,the feature information of the image are extracted using a variety of feature extraction methods.Finally,basing on the idea of the frame of information compression dictionary fusion mechanism proposed by Zhao,we propose the features fusion strategy of dictionary,and applied it into multiscale sparse representation classification method.By simulating SAR image and real SAR image experiment,the proposed classification model not only has strong robustness in suppressing noise,but also reduces the complexity of traditional algorithm.2)A multi-layer SAR image classification method based on superpixel is proposed.In view of the problems of the existing methods of processing the SAR image,such as high time consuming,and the weak edge classification ability.A new method is proposed based on the above research content 1).The method consists of three parts: superpixel segmentation,multiscale feature extraction,and hierarchical classifier construction.Classification of each layer is operated by SR classifier in hierarchical structure classifier.And we introduce threshold to divide the results into certain result and uncertain result.The certain results regard as the training samples of the next layer.This method effectively makes up the defects of samples in the sparse representation classification process.It not only implements the sparse representation classification,but also updates the dictionary.The experimental results show that this method is not only beneficial to the classification of small samples,but also has the advantages of fast convergence speed and the ability to deal with the classification of heterogeneous regions.