Hyperspectral Image Anomaly Detection Method By Fusing Spatial And Spectral Information

Hyperspectral image is a new type of remote sensing data,it can provide continuous dozens or even hundreds of bands of spectral information.Compared with multi-spectral images,hyperspectral images not only have very high spectral resolution,but also contain rich spatial information.Because the spectral information of different material is very dif-ferent,hyperspectral images provide the ability to detect materials with different spectral characteristics.Anomaly detection is one of the important research direction in the hyper-spectral target detection field.Anomaly detection does not require prior knowledge of the spectral information of the target,but on the basis of that whether there is a significant difference between the spectral characteristics of the pixel to be detected and its neighborhoods to determine whether the pixel belongs to an anomaly or background.Currently hyperspectral image anomaly detec-tion faces many challenges,such as the high volume of hyperspectral images data,the data redundancy between the spectral information is high,how to effectively combine the spatial information of the image with the spectral information,which are all urgent problems to be solved.Therefore,fast and efficient hyperspectral anomaly target detection algorithm is required and this is the main goal of the researchers.In this thesis,attention is paid to the characteristics of anomalous targets in hyperspectral images,and the rich spatial and spec-tral information of images are involved.Two hyperspectral image anomaly target detection algorithms are proposed based on the combination of the spatial and spectral information of images so as to obtain better detection results.The main research work and achievements of this paper can be summarized as follows:(1)An anomaly detection algorithm based on attribute filtering and edge preserving filtering is proposed,which makes full use of the distribution characteristics of background and anomaly targets in the hyperspectral image to determine the potential location of the abnormal target.With Boolean map as a penalty function,the algorithm can remove the interference of background information.Finally,with a real-time edge-preserving filter,the problem of low detection efficiency caused by limited using of spatial information has been well addressed.Experimental results show that compared with other widely used anomaly detection algorithms,the proposed method is able to obtain better results in rather short time.(2)An hyperspectral anomaly detection method via shape adaptive joint sparse repre-sentation is proposed.With the help of shape adaptive algorithm,the spatial information of hyperspectral image are discovered to find the local neighborhoods of each pixel,and then provide a selection range of pixels to make up the background dictionary.Joint sparse model is used to remove the interference caused by abnormal target and thus,maximizing the sep-arability of abnormal targets and background.Experimental result show that this method can relatively express and use well of the rich spectral and spatial information contained in the image.