Hyperspectral Target Detection Based On Sparse Representation

Hyperspectral image contains a wealth of spatial information and spectral information.Hyperspectral images have huge advantage in target detection area,which contributes to the rapid development of the technology of hyperspectral image target detection.Target detection using hyperspectral image possesses several advantages.Firstly,due to the large number of bands in hyperspectral image data,the spectral features used for target detection are more abundant and detailed.As a result,the ability to detect and identify targets is greatly improved,and even some subtle differences among objects can be distinguished.Secondly,hyperspectral remote sensing technology is capable of effectively extracting the radiation characteristic parameters of various objects,which greatly increases the success probability of quantitative analysis of the target features.Target detection using hyperspectral image has been successfully applied in military,civil and other fields.This paper mainly studies the related problems in hyperspectral image target detection,the main work and achievements of the paper are as follows:(1)The existing representative target detection methods of remote sensing image are introduced and analyzed.Based on the characteristics of hyperspectral images,concepts related to hyperspectral image target detection are introduced.Four methods of target detection in remote sensing images are described in details,and their advantages and disadvantages are analyzed?(2)Hyperspectral signal representation models are introduced and analyzed.Three commonly used hyperspectral signal representation models and the related hyperspectral image target detection algorithms are introduced.(3)A new hyperspectral image target detection algorithm is proposed,which incorporates sparse representation and traditional operator.Assume that the spectral data is distributed over a multidimensional space with probability,and the sample is used to make a reasonable estimate of the statistics in the distribution and perform correlation analysis.With the help of the architecture of traditional target detection algorithm,this paper proposes the target detection algorithm based on sparse representation coefficients,in which the dictionaries are constructed by using spectral immixing technique and unsupervised clustering technique respectively.The corresponding sparse representation coefficients are also obtained to complete the sparse representation.Detection decision is obtained by using the improved CEM detector,by comparing the calculation results with the predefined threshold.Experiments using real hyperspectral image data illustrate that the newly proposed incorporates sparse representation and CEM operator algorithm exhibits better detection performance than the representative traditional detection algorithms,as well as the representative detection algorithms based on sparse representation.