Study On Anomaly Detection Of Hyperspectral Image

Hyperspectral remote sensing image is a three-dimensional image data,which combines spectral and spatial information.Compared with single-band images,hyperspectral image contains more abundant spectral information.Hyperspectral image anomaly detection is an important research direction in hyperspectral image processing.It refers to the detection of anomaly targets in image when the spectral information of objects is unknown.At present,it is widely used in surveillance,environmental monitoring and so on.In view of the data characteristics of hyperspectral image and the shortcomings of existing anomaly detection algorithms,this thesis studies and implements the anomaly detection algorithm of hyperspectral image from two perspectives of extracting anomaly information from hyperspectral image and making full use of spatial information.An anomaly detection algorithm based on nonlinear dimension reduction and image fusion for hyperspectral image is studied.In view of the large amount of hyperspectral image data which is difficult to process,the algorithm performs nonlinear dimensionality reduction on the image and extracts anomaly information,then optimizes the detection performance of the algorithm through fusion processing.The algorithm uses the kernel principal component analysis and independent component analysis to reduce the dimensionality of the original hyperspectral image.By combining the non-linear dimension reduction with the linear dimension reduction algorithm,the difficulty of the direct processing of the hyperspectral image is reduced,and the information contained in the hyperspectral image can fully excavated.The joint skewness-kurtosis figure is used to further extract the anomaly component and make the anomaly component contain more anomaly information which is conducive to the processing of the following anomaly detection algorithms.The algorithm breaks through the traditional single processing method,and uses three algorithms to detect the anomaly component.Finally,the detection results are fused by the fusion algorithm based on the wavelet transform,which can improve the quality of the detection results.An anomaly detection algorithm for hyperspectral image based on spectral-spatial is studied.Extending anomaly detection to spatial dimension overcomes the disadvantage that traditional anomaly detection algorithm is limited to spectral dimension information,and can improve the detection performance.The algorithm based on collaborative representation,uses locally linear embedding dimension reduction and potential anomaly singnal to noise ratio to process the original hyperspectral image,and the obtained anomaly component is used to carry out the collaborative representation based detection of spatial dimension.The detection result can be used as posterior information to construct a hyperspectral image with approximate background,from which background dictionary is obtained.The background dictionary can assist in constructing the weight coefficients in the process of cooperative representation of spectral dimension.Then the final detection result of the algorithm is obtained.The algorithm uses the first detection result as a posterior information to assist the second anomaly detection,which greatly improves the accuracy of the algorithm.This paper evaluates the performance of the algorithm by using the ROC curve and the area under the curve.Two algorithms studied and implemented compare five typical algorithms separately and carry out anomaly detection simulation experiments with real hyperspectral image.The experimental results show that the two studied algorithms can highlight the target,suppress the background and have good anomaly detection performance.