Research On Synthetic Aperture Radar Images Vehicle Target Classification Based On Discriminative Dictionary Learning

Synthetic aperture radar?SAR?imaging technology breaks through the limitations of weather,light and other factors,which enables high-resolution imaging in all day and all weather.At present,SAR imaging technology has been widely applied in many areas such as national defense and military,geological exploration,ocean exploration,city planning and so on.SAR target classification is one of core technology for SAR applications,and it attracts considerable attention of researchers at home and abroad.It is no doubt that one of the keys to the SAR image target classification system is how to devise the classification algorithm.In recent years,a large number of representation learning algorithms like sparse representation method and dictionary learning method are introduced into the field of SAR target classification and obtained good performance.This thesis specializes the SAR target recognition method based on discriminative dictionary learning,and the detailed works in this thesis are as follows:?1?In traditional dictionary learning method,it usually uses ₀ or ₁ norm to restrict the sparse representation coefficient causing complicated operations and time consuming during training process.Besides,the global aspect SAR target images are considered to be equally correlated.According to the above issues,it is proposed that adaptive local aspect dictionary pair learning method for SAR target images classification.The proposed method utilizes non-negative sparse learning method to select the local aspect sector of the query sample adaptively and learns a dictionary pair from the selected local aspect sector jointly.During the training dictionary process,the analysis dictionary is introduced to project the coding matrix linearly,which avoids ₁norm sparse coding,improves the training efficiency of the algorithm and reduces the calculation cost.The experimental results of the proposed method show that it reduces the interference of unrelated samples and promotes discriminative power between different targets effectively.And the accuracy classification rates and noise robustness are further improved.?2?In the label consistent K-singular value decomposition?LC-KSVD?method,it restricts the same class samples with similar sparse coding forms,which discards the strong correlation of SAR images among local aspects.According to this problem,the method united extreme learning machine method and aspect label consistent K-singular value decomposition dictionary learning for SAR image target recognition is proposed.The method adopts extreme learning machine to learn the local aspect characteristics of the testing samples,and learns a dictionary which takes account of discriminability and reconstruction ability from the aspect consistent training sample subset.Furthermore,the dictionary learning model makes the most of the label and local aspect information to promote the classification performance of the method.The experimental results verify the effectiveness of the proposed method.Compared with the basic LC-KSVD method,the classification performance of the proposed method is improved.