Hyperspectral anomaly detection based on generative adversarial network

Abstract

The invention provides a hyperspectral anomaly detection method based on a generative adversarial network. The method includes the steps of first, constructing a network model that conforms to the characteristics of hyperspectral image data; second, training, by using a hyperspectral data set as a training sample, a probabilistic model that conforms to the statistical characteristics of hyperspectral data; third, calculating the probability that each pixel to be detected belongs to the real hyperspectral data set by using a trained discriminator; and fourth, calculating an anomaly detection result for each pixel in the hyperspectral data set. In the invention, the idea of the generative adversarial network is used to train the probabilistic model conforming to the characteristics of hyperspectral image data, the model can generate the generative data set that conforms to the overall statistical characteristics of the real hyperspectral data set and can judge the probability that one pixel belongs to the hyperspectral data set. The method of the invention does not require model assumption of the characteristics of the hyperspectral data set and can directly estimate the probabilistic model that conforms to the statistical characteristics of the hyperspectral data to be detected, so that the detection results are more reliable.

Description

technical field [0001] The invention relates to a hyperspectral image abnormality detection method. Background technique [0002] Hyperspectral remote sensing images have rich spectral information. It has great advantages in the fields of target recognition, classification, and extraction of interested target information. Hyperspectral anomaly detection is an unsupervised method that does not require prior information on the target and background, and can find out abnormal targets with obvious deviations between different spectra and the surrounding background. In most cases, prior information of objects and backgrounds is difficult to obtain, so unsupervised anomaly detection techniques are of great significance in practical applications. [0003] In traditional RX anomaly detection, it is usually assumed that the features of hyperspectral data obey the Gaussian normal distribution, and then the probability of the pixel belonging to the abnormal point is estimated by cons...

AI Technical Summary

Problems solved by technology

However, due to the high-dimensional nature of hyperspectral data, the characteristics of hyperspectral data sets usually do not conform to any ideal mathematical distribution, and it is difficult for the Gaussian normal distribution to effectively estimate the distribution of hyperspectral data, resulting in accurate anomaly detection results. low degree

Images