Research On Hyperspectral Anomaly Target Detection Algorithm Combining Space And Spectral Information

Hyperspectral remote sensing combines imaging technology with spectral technology,which can provide more abundant information for earth observation tasks.At present,the mainstream anomaly detection methods only take spectral differences as the core basis and cannot fully mine the information contained in hyperspectral images.Therefore,this paper studies the anomaly detection algorithm of hyperspectral images by combining spatial and spectral information.Starting with the study of the correlation between adjacent bands and pixels of hyperspectral data,this paper analyzes the problems existing in the traditional representation of hyperspectral images,and demonstrates the necessity of spectral dimension data compression,space spectrum information integration and tensor theory.The main research contents of this paper are as follows:(1)Aiming at the problem that the anomaly detection algorithm based on local window misjudged the global non-abnormal edge information as local anomaly,this paper proposes a local anomaly detection algorithm based on linear background removal.Firstly,principal component analysis was carried out to detect the edge of each extracted principal component,and the maximum value of the edge detection operator of each principal component was taken as the final image edge.Then,line information is extracted from edge images by Hough transform.Finally,the improved method is used to detect anomalies in hyperspectral images.The improved method and the original method are implemented under different double window sizes.The comparative experiments show that the improved algorithm is less sensitive to the window size,and the local false alarm is effectively reduced.(2)Aiming at the problem that the assumption that the background statistical characteristics obey the Gaussian distribution of classical anomaly detection algorithms such as RX is not always valid,the image Gaussivity is improved based on neighborhood weighted residual image estimation.The algorithm uses the neighborhood spectral information to estimate the center pixel,and the residual image obtained by the difference between the estimated image and the original image is used for anomaly detection.The experimental results show that the improved method can improve the detection performance obviously compared with the original method,and it also introduces the neighborhood spatial information and improves the utilization rate of hyperspectral data.(3)Aiming at the shortcomings of common data representation methods of hyperspectral images and the problems of target and noise pollution in background estimation,a hyperspectral anomaly detection algorithm based on tensor decomposition background purification was proposed.Firstly,the image to be processed is represented by the third-order tensor and the whole Tucker decomposition is performed.Then,principal component analysis was carried out on the decomposed factors of the three directions respectively,and the abnormal noise part was reconstructed from the secondary part of each factor and the corresponding sub-tensor.The optimal principal component combination of the joint anomaly degree of the empty spectrum of the abnormal and noise part reached the peak through cyclic solution.Finally,a relatively pure background information is obtained to eliminate the abnormal noise.The experimental results show that the improved algorithm can effectively reduce the false alarm rate and increase the separation degree between the target and the background.