The Research Of The Tensorial Independent Component Analysis

Polarimetric SAR images contain rich target information which can be use for the target detection,classification,recognition and other areas.And those target information need to rely on polarimetric features to be found and applied.Along with the development of Pol SAR theories and Pol SAR system,a large number of polarimetric features can be found and collected.Since each feature is just a simple description of the polarization scattering characteristics of a target.So it is unable to distinguish different types of objects which have similar scattering characteristics or the same type of target with different scattering characteristics.Therefore,in order to carry out the classification and identification of target better,it have to combine a variety of polarimetric features.Although the combination of multiple polarimetric features can collect more useful information from the target,it also adds a lot of redundant information which will not only greatly improve the computational complexity but also reduce the classification accuracy of the final experiment.At the same time,due to a large number of speckle noise exist in the Pol SAR image,the polarimetric features obtained from Pol SAR image will contain a lot of noise.If we use these polarimetric features in classification simply,we will get the worse classification results.The existing algorithm which is the independent component analysis based on tensor decomposition(TICA)is an algorithm to classify the polarimetric SAR images by combining various polarization characteristics.So it must face these disadvantages.In view of the above problems,this paper will improve the existing method TICA in order to get better polarization SAR image classification effect.Its main contents are as follows:(1)A Pol SAR image classification algorithm which improves traditional method TICA by weighted nucler norm minimization(WNNM)is proposed.We uses the weighted nucler norm minimization for the preprocessing of the input data to improve the traditional TICA algorithm.By the end of the classification results,we can know that this algorithm can effectively improve the accuracy of classification of Pol SAR images.(2)A Pol SAR image classification algorithm which improves the Fast Independent Component Analysis(Fast ICA)of the traditional method TICA is proposed.In this improved algorithm,we improve the Fast ICA of the traditional TICA algorithm to remove data redundancy which is due to the combination of a variety of polarimetric features.By the classification results,we can know that this improved algorithm can effectively improves the effect of Pol SAR images classification.(3)According to the advantage of the improved algorithm before two,a new improved TICA algorithm is proposed.In this improved algorithm,That is to say we uses the weighted nucler norm minimization and the improved Fast ICA to improve the TICA algorithm at the same time.Through the result of the classification,it is well known that this algorithm is better for the classification of the Pol SAR images.