Polarimetric SAR Image Classification Based On Spatial Information And LSTM Network

As a high precision imaging radar,polarimetric synthetic aperture radar(Pol SAR)transmits electromagnetic waves with strong penetration and plays an important role in military affairs,industry and agriculture.Therefore,the research on polarimetric SAR image interpretation has gradually developed.The classification of polarimetric SAR is a major branch in the field of remote sensing image interpretation.Traditional polarimetric SAR classification methods depend on the coherence characteristics of polarimetric SAR data,but without the use of spatial information of images,they can not achieve good classification results.The main work of this paper is to classify polarimetric SAR images based on spatial information and long short term memory network.Based on deep confidence network and long short term memory network,it is applied to polarimetric SAR classification.The main work of this paper is as follows:1.A polarimetric SAR classification method based on depth confidence network of image spatial information is studied.First,the polarization data to be classified are pre-processed and their coherence matrix is obtained.Then,the depth confidence network is trained by using the coherence matrix to extract the characteristics of polarization SAR data.Then,the polarization SAR data is preliminarily classified through the softmax layer.Finally,the local spatial information of the data is introduced to optimize the initial classification results.Good classification results are obtained.2.A polarimetric SAR classification method based on SAE-LSTM network is studied.Because polarimetric SAR data contain abundant polarimetric scattering information,only processing polarimetric coherence matrix is not enough to characterize the complete terrain information.In this paper,firstly,polarimetric SAR data are processed by multiple target decomposition methods to obtain its multi-dimensional eigenvectors.Then,SAE network is used to select the features of the obtained multi-dimensional eigenvectors to obtain the tribute of classification effect.The characteristics of high devotion.Finally,considering the spatial and temporal information of polarimetric SAR data,LSTM network is used to classify polarimetric SAR images.3.A classification method of SAE-LSTM polarimetric SAR based on super-pixel segmen-tation is studied.Firstly,SAE network is used to select feature vectors for classification.Secondly,the super-pixel segmentation algorithm based on feature decomposition is used to generate super-pixel blocks.Finally,the LSTM network's initial classification results are compared with the super-pixel blocks,and the prediction labels corresponding to the positions are corrected according to the labels of the super-pixel blocks,and the final classification results are obtained.