Research On SAR Image Classification Algorithm Based On Bayesian Sparse Representation Theory

Synthetic Aperture Rader(S AR)is a microwave imaging radar with many unique advantages,such as being exposed to light and weather,enabling observations throughout the day,and even obtaining information that is obscured by other objects.With the development of science and technology and the advancement of the times,the use of SAR imaging technology has become more and more extensive.SAR image classification and detection have many application prospects in military and civilian applications,and have gradually become the main research direction of SAR image interpretation.Based on sparse representation theory,Bayesian sparse representation,spatial semantics,sub-categorization concepts,etc.,a series of research work on classification and detection techniques for typical SAR image ground targets(vehicles,buildings,airports,etc.)is carried out.as follows:(1)A sub-classification scheme of SAR image training set based on K-means clustering algorithm is designed.In the existing sub-category classification model,the sub-category algorithm is only for optical images,and the feature distinguishing ability of the target is strong.Due to the particularity of its imaging,SAR images contain less information,low signal-to-noise ratio and some speckle noise,which leads to large changes in the representation of targets in SAR images,weaker discrimination,and speckle noise.The accurate description of the features causes interference,which causes certain intra-class differences and inter-class blurring of SAR image targets.In this paper,based on the characteristics of SAR images,a sub-classification scheme of SAR image training set is designed.By further sub-categorization of training set categories,the intra-class differences of images are reduced,the performance of classification models is improved,and SAR image targets are reduced due to intra-class differences.The error caused.(2)A SAR image target classification model based on Bayesian sparse representation is constructed.Aiming at the problem that the sparse representation is easy to fall into the local optimal solution during the solution process,a Bayesian sparse representation is introduced to approximate the sparse representation,so that the test sample data can be solved based on the sparse representation of the training set dictionary.The most sparse solution enables the sparse reconstruction of subsequent samples to achieve better accuracy.Then,based on the K-means clustering algorithm,the training set sub-classification scheme is combined with the Bayesian sparse representation framework to construct a classification model based on Bayesian sparse representation,and after sparse reconstruction of multiple features of the test sample.The reconstructed residuals are linearly fused,which reduces the misclassification caused by the ambiguity between SAR images and improves the classification performance.Experiments based on real data sets confirm the feasibility of the proposed method.(3)A SAR image airport runway detection method based on Bayesian sparse representation fusion framework and spatial semantic matching is proposed.The algorithm mainly includes two parts:runway evaluation and confirmation.Firstly,the Bayesian sparse representation framework and linear fusion are combined to formulate the runway evaluation strategy of the airport main runway(PR)and the auxiliary runway(AR)for the region of interest(RoI).Conduct an airport assessment and enter the validation process if the assessment criteria are met.Then,the spatial semantic relationship is included in the final confirmation procedure for runway detection,based on the spatial relationship between the airport main runway(PR),the auxiliary runway(AR)and the entire runway area(RR)runway,based on its directional gradient histogram(HoG).Two spatial semantic rules are developed to verify each candidate's region of interest.If the match is successful,the candidate's region of interest is identified as an airport target.Since the fusion result is determined by multiple reconstruction residuals,a robust evaluation result of PR and AR can be obtained.By developing the airport main runway(PR),the spatial relationship between the auxiliary runway(AR)and the entire runway area(RR),the proposed method produces higher detection accuracy than other solutions.Test results based on real scenes demonstrate the effectiveness of our approach.