PolSAR Image Classification Based On Markov Discriminative Spectral Clustering

PolSAR image classification is an important research content of remote sensing image interpretation,and also a key step in the application of remote sensing image processing theory.Through classification,it can verify the correctness of the basic theoretical algorithm,and can be applied to other fields such as change detection and target recognition.The spectral clustering method is characterized by simple and fast clustering analysis of complex data structure information.Therefore,how to use spectral clustering method to classify PolSAR images quickly and accurately has important research significance.In this paper,the classification method of PolSAR image based on Markov discriminative spectral clustering is proposed firstly.Then,based on the algorithm,the characteristics of PolSAR data are improved in multi-view and large-scale.The main research contents of this paper are as follows:A.Aiming at the low accuracy of existing spectral clustering methods for PolSAR image classification,this paper proposes a Markov-based discriminative spectral clustering method(MDSC)with low rank and sparse decomposition.Firstly,a real low-rank probability transfer matrix is restored as an input to the standard Markov spectral clustering method to reduce the influence of noise on the classification result.Then the discriminative information is introduced into the objective function to make the PolSAR image.The data information can be used in multiple levels.Finally,the augmented Lagrangian multiplier method is used to solve the objective function optimization problem under low rank and probability simplex constraints.Experiments on different data sets show that the method has good accuracy and low sensitivity,showing good classification performance.B.Because the ordinary single-view spectral clustering method can not effectively use other information of PolSAR images,this paper proposes a multi-view discriminative spectral clustering algorithm based on Markov discriminative spectral clustering method.The algorithm extracts the polarization target decomposition features and texture features as two perspectives of the PolSAR image,and then performs LDA cross-dimension reduction on two views at the same time.Finally,the Markov discriminative spectral clustering method is used for clustering and alternate iteration.Classes of different perspectives converge until the final classification results.The experimental results show that the proposed algorithm significantly improves the clustering performance,and the effectiveness of the algorithm is verified by the actual data set.C.In order to make up for the application bottleneck caused by the high computational complexity of spectral clustering algorithm,this paper proposes a large-scale discriminative spectrum clustering algorithm based on feature selection based on Markov discriminative spectral clustering method.Firstly,the algorithm uses clustering forest to select the features of PolSAR data,and selects the best combination of feature sets as the initial clustering vector of feature growth,which effectively reduces the feature dimension and suppresses noise.Secondly,it passes the PolSAR image.The super-pixel segmentation selects the representative points,and designs a similarity matrix which saves memory and is easy to calculate.Finally,the clustering result is obtained by Markov discriminative spectral clustering algorithm.Experimental results on different data sets show that the algorithm does achieve better classification accuracy on large-scale PolSAR images.