A hyperspectral image anomaly detection method based on constrained sparse representation

Abstract

The invention belongs to the field of image processing, and relates to a hyperspectral image anomaly detection method based on constrained sparse representation, comprising the following steps: (S1) linearly normalizing the hyperspectral image; (S2) for each test pixel, according to the double-window The model extracts the local background dictionary; (S3) Solve the constrained sparse representation model according to the local background dictionary to obtain the optimal solution 1 of the model; (S4) Delete all abnormal atoms from the local background dictionary according to the optimal solution 1 of the model to obtain New background dictionary; (S5) According to the new background dictionary, solve the sparse representation model with constraints to obtain the model optimal solution 2; (S6) calculate the detection value of the pixel according to the model optimal solution 2; (S7) traverse the entire For hyperspectral images, the detection value is calculated for each pixel of the hyperspectral image, and the image composed of these detection values ​​is output, that is, the anomaly detection image. The invention does not need background statistical information and setting sparsity, and improves the reconstruction accuracy.

Description

technical field [0001] The invention belongs to the field of image processing, and relates to a hyperspectral image anomaly detection method based on sparse representation with constraints. Background technique [0002] Hyperspectral image target detection is one of the important directions of hyperspectral remote sensing applications, covering many fields such as environmental detection, urban survey, mineral mapping and military reconnaissance. Anomaly detection does not require any target spectral information and has a broader application prospect. In anomaly detection, outliers are often detected by testing the significant difference between the pixel spectrum and its local background spectrum. [0003] The classic anomaly detection algorithms include Reed-Xiaoli (RX) detector, Kernel RX detector, support vector data description (SVDD) detector and so on. However, when the local background contains target signals, the performance of traditional anomaly detection algori...

AI Technical Summary

Problems solved by technology

[0004] In view of the deficiencies of the existing technology, in order to solve the above-mentioned problem that the local background is polluted by the target signal, and the problem that the degree of sparsity in the sparse representation is difficult to determine, the specific technical solutions adopted by the present invention are as follows

Images