Reasearch On Hyperspectral Anomaly Detection Method Based On Low Rank Sparse Decomposition And Super-resolution Processing

With the development of remote sensing technology,the spectral resolution of hyperspectral image is increasing,which makes hyperspectral image become the research hotspot in the field of remote sensing.In target detection,according to whether the prior spectral information of anomaly target can be obtained,there are two types: supervised and unsupervised.Anomaly target detection belongs to the category of unsupervised target detection,which does not need the prior spectral information of anomaly targets.It is difficult to obtain the real spectral information of the ground target,so anomaly detection is more practical.At present,the d hyperspectral anomaly detection faces many challenges: the one is how to avoid discussing the distribution of background data,the two is how to effectively combine the spatial information and spectral information of hyperspectral image,and another is how to detect the anomaly targets by the phenomenon of mixing pixels caused by the lower spatial resolution and higher spectral resolution in hyperspectral image Therefore,the research of fast and efficient detection method of hyperspectral anomaly detection is the focus of this paper.In this paper,the data characteristics of hyperspectral image and how to effectively combine the spectral information and spatial information of hyperspectral image are studied.The main research work and achievement are as follows:Firstly,this paper briefly introduces the imaging mechanism of hyperspectral image,analyzes the data characteristics of hyperspectral image in detail,and focuses on the anomaly detection theory and classical detection methods of hyperspectral image and the problems in anomaly detection.In addition,the experimental data used in hyperspectral anomaly detection experiment and the performance evaluation method and evaluation index of anomaly target detection are introduced,which lays the foundation for the later research work.Secondly,because the distribution of background data in hyperspectral image is very complex,and the assumption of background distribution is idealized in classical detection methods,which leads to low detection rate and high false alarm rate,an anomaly detection method of hyperspectral image based on low-rank sparse decomposition and correlation constraints is proposed.Firstly,the low-rank sparse decomposition of hyperspectral image is realized by tensor analysis and self-updating sparse representation,to suppress the background part and noise part.Then,the anomaly detection of the abnormal target is realized by combining the spatial correlation constraint and the object correlation constraint.The algorithm can avoid the discussion of background data distribution,achieve the maximum suppression of background and noise,and enhance the anomaly target.Finally,aiming at the low spatial resolution of hyperspectral image,the mixed pixel phenomenon caused by the high spectral resolution and the same object and different spectrum phenomenon produced in the imaging process,which affect the anomaly detection of small target and sub-pixel target,aiming at the anomaly detection of small target and sub-pixel target,the hyperspectral anomaly target based on super-resolution processing is proposed.The spectral variability is considered by the phenomenon of the same substance and different spectrum,and the method of unmixing based on the extended linear mixture model is introduced.The highprecision unmixed component image is obtained,and then the spatial resolution of hyperspectral image is improved by super-resolution processing.Finally,the anomaly detection is completed by combining LRX and ground object correlation.This method solves the problem of low spatial resolution of hyperspectral image,overcomes the interference caused by mixed pixels and the phenomenon of the same object and different spectrum in the image to the detection of small and sub-pixel targets,highlights the anomaly targets and improves the accuracy of hyperspectral anomaly detection.