Polarimetric SAR Image Classification Based On Sparse Learning And Metric Learning Classifier

Polarimetric Synthetic Aperture Radar is widely used,and one of the most important applications is image classification.Compared with the traditional Synthetic Aperture Radar system,images obtained by the Polarimetric SAR contain more abundant features information and the feature dimension is higher.If we want to popularize classification methods of the Polarimetric SAR images in real life,first of all,the methods must be fast and accurate.In fact,the key factors that influence the classification accuracy and classification complexity are feature extraction and classifier design.However,the performance of the classifier depends on features.Some traditional algorithms have a misunderstanding,that is,the more features there be,the higher accuracy of classification.And the traditional algorithms are usually based on pixels,which ignored the spatial information between the pixels.In order to solve the problems that the classification accuracies of some traditional methods are not high enough,and the extraction of features cost too much time,we will introduce three kinds of algorithms.The main work focuses on the use of the deep features of Polarimetric SAR images.Introduce the idea of super-pixels,and combine the deep features with metric learning classifier.The work that we have done is as follows:We proposed a Polarimetric SAR feature classification method based on the metric learning classifier.It is different from the traditional feature classification algorithm,which introduces the framework of the metric learning classifier.It can call the existing support vector machine algorithm model,which can effectively eliminate invalid information and redundancy information to get higher classification accuracy.Finally,compared with the results of three groups of experiments,we can verify the feasibility and effectiveness of the algorithm.A Polarimetric SAR feature classification method based on Sparse Auto Encoder and metric learning classifier is proposed.Combine the deep features with metric learning classifier.This method uses a stack of sparse auto encoder to extract the deep features of Polarimetric SAR images and uses the metric learning classifier to classify.Compared with three experimental results,the classification performances are better and the accuracies are higher.We proposed a Polarimetric SAR feature classification method based on super pixels and metric learning classifier.Many common methods are based on pixels.This algorithm takes the spatial correlation between pixels into account and adds the idea of super pixels.By combining super pixels with the metric learning classifier,not only the spatial continuity of the Polarimetric SAR image is maintained,but also the classification accuracy is ensured.Compared with three experimental results,the classification performances are better and the accuracies are higher.