High-spectrum abnormal object detection method based on background dictionary learning and structure sparse expression

Abstract

The invention discloses a high-spectrum abnormal object detection method based on background dictionary learning and structure sparse expression, for solving the technical problem of low object detection efficiency by use of a conventional high-spectrum abnormal object detection method. The technical scheme is as follows: after an initial background pixel is selected based on a local RX algorithm, a robust background dictionary is obtained through learning by use of a principle component analysis dictionary learning method. In a sparse vector solving and image reconstructing process, heavy-weight Laplace prior is introduced, and thus the sparse vector solving precision is improved. Finally, according to errors between an original image and a reconstructed image, accurate extraction of an abnormal object is realized. Test results on a real high-spectrum satellite image AVIRIS and a simulated high-spectrum data set show that the detection rate of a detection result obtained by use of the method is improved by 8% to 15% under the condition of a constant fault alarm rate compared to background arts.

Description

technical field [0001] The invention relates to a hyperspectral abnormal target detection method, in particular to a hyperspectral abnormal target detection method based on background dictionary learning and structural sparse representation. Background technique [0002] Hyperspectral anomaly target detection technology is a hyperspectral target detection technology that does not need to provide prior spectral information of the target to be detected, and has strong practicability in practical applications. [0003] In the traditional abnormal target detection algorithm, it is assumed that the abnormal target in the hyperspectral image data is a small probability event relative to the image background, and the global or local statistical characteristics can be used to detect the abnormal target. This type of algorithm generally assumes that the image background has statistical characteristics that obey the Gaussian distribution, but in practical applications, due to the limi...

AI Technical Summary

Problems solved by technology

[0005] In order to overcome the shortcomings of low detection efficiency of the existing hyperspectral anomaly target detection methods, the present invention provides a hyperspectral anomaly target detection method based on background dictionary learning and structural sparse representation

Images