Research On Hyperspectral Image Anomaly Detection Algorithm

With the development of science and technology,remote sensing imaging technology has gradually developed from multispectral imaging to hyperspectral imaging.Compared with multispectral images,hyperspectral images have higher spectral resolution,so that different objects that are difficult to be distinguished in multispectral images can be better distinguished by hyperspectral images.Compared with target matching detection,hyperspectral anomaly detection does not need a priori information of ground objects,and can automatically complete the detection process according to the difference between the target and the surrounding background.Therefore,it has broad application prospects in searching for mineral resources,observing and collecting meteorological data,and monitoring the health of vegetation.Based on the collaborative representation theory and the low-dimensional manifold modeling theory,the anomaly detection algorithm of hyperspectral image based on collaborative representation and the anomaly detection algorithm of hyperspectral image based on the low-dimensional manifold modeling are studied in this paper.In order to solve the problem that abnormal pixels polluting the background dictionary in the algorithm of hyperspectral image anomaly detection based on collaborative representation,a hyperspectral anomaly detection algorithm based on low-dimensional manifold modeling and collaborative representation is researched and implemented in this paper.Firstly,the algorithm uses low-dimensional manifold modeling to reconstruct the background image without anomalies from the multiple random sampling of the original hyperspectral image,and then uses the background image to construct the background dictionary required for collaborative representation.Finally,the collaborative representation residuals of the original image under the background dictionary are calculated to realize anomaly detection.In this paper,low-dimensional manifold modeling is used to improve the anomaly detection algorithm of hyperspectral images based on collaborative representation.By eliminating the potential abnormal targets in the background dictionary,the detection performance of the collaborative representation anomaly detector is improved.Aiming at the problem that there are many redundant calculations in the anomaly detection algorithm of hyperspectral image based on low-dimensional manifold modeling,this paper studies and implements the anomaly detection algorithm of hyperspectral image based on stack autoencoder and low-dimensional manifold modeling.In view of the fact that the stacked autoencoder can automatically extract the in-depth feature information of the original data according to the task requirements,the algorithm obtains the reduced-dimensional image of the hyperspectral image by constraining the number of coding units of the stacked autoencoder to decrease layer by layer,and then uses the low-dimensional stream The low-dimensional anomaly detector performs anomaly detection on the reduced-dimensional image and obtains the detection result.In this paper,a stacked autoencoder is used to reduce the redundant calculation in the low-dimensional manifold anomaly detector,and at the same time improves the detection performance of the detector.In order to verify the effectiveness of the algorithm in this paper,representative algorithms based on different background assumptions were selected as the comparison algorithm.Simulation experiments were carried out on the synthetic hyperspectral image data set and the real hyperspectral data set.A variety of evaluation methods were used to compare and analyze the detection performance of the algorithm in this paper.The results of simulation and comparison experiments show that the algorithm studied in this paper has better detection performance compared with the comparison algorithm.