High Resolution Hyperspectral Remote Sensing Image Target Detection

With the continuous development and progress of remote sensin g imaging technology,remote sensing images are moving towards higher temporal resolution,higher spatial resolution and higher spectral resolution.In this case,the application requirements of remote sensing images are increasing,and remote sensing imag es are applied in more and more different areas.At the same time,it also brings more and more problems and difficulties that cannot satisfy the application requirement for remote sensing data processing.For high resolution hyperspectral images,the higher spatial resolution brings us more details about the objects on earth,it will also produce richer spectral characteristics,correspondingly.But it also brings more difficulties for traditional target detection methods.Therefore,how to exploit the spatial information of the high resolution hyperspectral data efficiently,is the key to solve the problems in high resolution hyperspectral image target detection.In this paper,starting from the characteristics of high resolution hyperspectral images,which possesses higher spatial resolution,we aim to dig the spatial and spectral joint information of high resolution hyperspectral images,and promote the performance of high resolution hyperspectral image target detection.First,we research the basic theories about tensor and superpixel segmentation technology.And then,the tensor target representation model and superpixel sparse representation model are studied.Finally,two different target detection methods are produced based on the tensor target representation model and superpixel sparse representation model.Compared with some conventional target detection algorithms,all of the proposed algorithms can achieve better results.The main work of this paper is to study the basic theories about tensor and superpixel segmentation,and produce the tensor target representation model and superpixel sparse representation of hyperspectral images to form the corresponding target detection algorithms,which contains the following three aspects:First of all,starting from the point of jointly exploiting the high resolution and spectral information in high resolution hyperspectral image,the tensor representation model and superpixel sparse model of hyperspectral image are studied.And this is the basic of proposing the target detection methods which jointly exploit the spatial and spectral information.Then,a subspace target detection algorithm based on local block tensor representation is proposed for the hyperspectral image target detection problem under high spatial resolution.The local block tensor representation is used to construct the target and the background projection subspace,and the tensor form of the generalized likelihood ratio detection operator,and achieved good detection results.Finally,aiming at the hyperspectral image target detection problem under high spatial resolution,a target detection algorithm based on the combination of super-pixel segmentation and sparse representation is proposed.Super-pixel is used as spatial constraint unit to realize sparse target detection of hyperspectral image.The simulation results show that the algorithm is superior to the existing sparse representation detection algorithm.