Hyperspectral image anomaly detection method based on joint extraction of spatial-spectral characteristics

Abstract

The invention discloses a hyperspectral image anomaly detection method based on joint extraction of spatial-spectral characteristics, and mainly solves the problem that in the prior art, there are many missed detection anomaly points. The method comprises the following specific steps: (1) constructing a deep belief network; (2) generating a hyperspectral training set; (3) training a deep belief network; (4) extracting a feature weight matrix and a bias matrix; (5) calculating the dimensional characteristics of each spectral vector in the hyperspectral training set; (6) detecting an abnormal value of the spectral vector dimension of the hyperspectral training set; (7) obtaining a spatial feature image of the hyperspectral training set; (7) obtaining a spatial feature image of the hyperspectral training set; and (8) obtaining an abnormal value of the hyperspectral image with the spatial spectrum characteristic. The hyperspectral image detection method can extract spectral characteristicsand spatial characteristics, can better distinguish abnormity and complex backgrounds in the hyperspectral image, and has the advantages of less detection result false detection abnormity and less detection result missed detection abnormity.

Description

technical field [0001] The invention belongs to the technical field of image processing, and further relates to a hyperspectral image anomaly detection method based on joint extraction of spatial spectral features in the technical field of image anomaly detection. The invention can be used to detect abnormal point targets and small area targets between the hyperspectral image and the background. Background technique [0002] Hyperspectral image outliers are pixels that differ greatly from the background spectral characteristic curve, and anomalous small areas are small areas that are spatially different from the background. Anomalies could be, for example, rare vegetation species, unusually growing vegetation, illegal plants associated with the drug trade, polluted areas of coastal waters, missing adventurers in the desert, buried archaeological structures, illegal border crossings and military vehicles under vegetation cover, seascapes Ships in the background and tanks in ...

AI Technical Summary

Problems solved by technology

Although this method can improve the accuracy of anomaly detection by solving the non-negative sparse matrix of the spectral image, the disadvantage of this method is that it needs to solve the sparse matrix of the hyperspectral image. Due to the huge data of the hyperspectral image, It makes the calculation process complicated and cumbersome, and at the same time, the method needs to perform threshold segmentation, which will introduce more human subjective factors, and make the method unable to process hyperspectral data obtained by different remote sensors with high precision.

Although this method can improve the effect of anomaly detection by using the cooperation between pixel spectra, the disadvantage of this method is that due to the complex background of hyperspectral images and more interference information, the anomaly detection method is directly applied to the background Hyperspectral images with complex and small-area anomalies can easily misdetect the background as an anomaly, resulting in low accuracy of anomaly detection

Although this method utilizes the spatial information of the hyperspectral image to achieve a good detection effect on small-area abnormal targets, the disadvantage of this method is that because the method needs to reduce the dimensionality of the hyperspectral image, the calculation The process is complicated and cumbersome, and only the spatial information is considered, which makes the detection performance of abnormal small target points worse, resulting in low accuracy of abnormal detection

Images