Research On Anomaly Target Detection Technology Of Hyperspectral Image Based On Sea Background

Hyperspectral images are widely used in military,agricultural,marine,and Geological exploration because of their high resolution,uniform spectrum,and large number of spectral bands.As one of the important applications of hyperspectral images,abnormal target detection has been a hot spot for experts and scholars.However,due to the huge amount hyperspectral data,the processing of the matrix data is difficult,and the complex background and the influence of various noises in the detection of abnormal targets make the detection of abnormal targets of hyperspectral images have great challenges.This paper mainly focuses on the problem of abnormal target detection of hyperspectral images based sea background,introduces the hyperspectral image simulation of sea background and the analysis of spectral characteristics of target and background,and proposes sparse representation and low rank and sparse representation for hyperspectral image anomaly detection.The main research contents of the thesis are as follows:The second chapter studies the simulation and characterization of hyperspectral image based on sea background.Since the hyperspectral image dataset based on the sea background is small and difficult to acquire at this stage,the infrared image acquisition of the sea background is simple.We introduce a method based on infrared image of sea background and use Gaussian model to simulate the whole hyperspectral image.Experiments and comparisons are carried out to prove the usability of the simulation method.Finally,we introduced the concepts of radiance difference,contrast and spectral angle,and analyzed the spectral characteristics of the target and background.The third chapter studies the abnormal target detection algorithm of hyperspectral image based on kernel sparse representation.The key to the sparse representation model is that the background of the hyperspectral image and the anomalous target are located in different subspaces.The dictionary composed of background elements is difficult to represent the abnormal target,so we can detect the abnormal target.Based on the sparse representation model,we further add the kernel method and the "sum to 1" constraint to achieve better performance on the actual hyperspectral dataset and the simulated sea background dataset.In the fourth chapter,the algorithm of hyperspectral image anomaly target detection based on low rank and sparse representation model is studied.Mainly based on the hyperspectral image matrix can be divided into a sum of a sparse matrix and a low rank matrix,and the low rank matrix can be used for abnormal target detection.Because the general algorithm tends to ignore the global structure in the process of abnormal target detection,the low rank and sparse representation model is based on the global matrix transformation,which makes some of the background points that are considered abnormal in the local can be better eliminated.In order to verify the performance of the algorithm,we used the measured hyperspectral dataset and the simulated dataset to experiment and achieved the desired results.