Hyperspectral anomaly detection method and system based on manifold constraint self-coding network

Abstract

The invention discloses a hyperspectral anomaly detection method and system based on a manifold constraint self-coding network, for solving the problem of low sample recognition rate. The hyperspectral anomaly detection method comprises the following implementation steps: (1) learning low-dimensional embedding manifold by adopting a manifold learning algorithm; (2) extracting hidden layer expression characteristics through a low-dimensional embedded manifold constrained self-coding network; (3) obtaining a global reconstruction error by using a low-dimensional embedded manifold constrained self-coding network; (4) calculating a local reconstruction error on the low-dimensional embedded manifold; (5) combining the global reconstruction error and the local reconstruction error to carry out abnormal target detection; and (6) counting experiment results, and calculating the precision of hyperspectral anomaly detection. According to the hyperspectral anomaly detection method, when the depthfeature of the hyperspectral image is extracted by using the self-coding network, the local feature of the image is considered; and the advantages of a global method and a local method are integrated, and a global reconstruction error and a local reconstruction error are considered at the same time during abnormity judgment, so that the abnormity detection precision is improved.

Description

technical field [0001] The invention belongs to the technical field of remote sensing information processing, and in particular relates to a hyperspectral anomaly detection method, which can be used in the fields of environment monitoring, mineral resource exploration, national security and the like. Background technique [0002] Abnormal targets in hyperspectral images generally refer to targets whose spectral information is very different from that of other surrounding pixels. In general, the abnormal targets in hyperspectral images are small targets and are sparsely distributed in the image. The biggest difference between hyperspectral anomaly detection and hyperspectral target detection is that hyperspectral anomaly detection does not need to know the prior knowledge of any target, that is, it does not need to know the spectral information of the abnormal target, and only needs to compare the target with the spectral characteristics of the surrounding environment. If th...

AI Technical Summary

Problems solved by technology

[0004] Although the above hyperspectral anomaly detection algorithms based on deep learning have achieved relatively good experimental results, they have not considered a problem: how many dimensions can reflect the hyperspectral data for the hidden layer features extracted by the self-encoder network or deep belief network essential characteristics? Moreover, since the abnormal target occupies a small proportion of the entire hyperspectral image, the hyperspectral anomaly detection method based on the autoencoder network or the deep belief network, in the process of model training, the background sample provides the main contribution to the training model, that is, based on The anomaly detection method of deep learning is essentially to learn the background of the target, and then detect the abnormal target

However, the existing hyperspectral anomaly detection methods based on deep learning do not consider the local characteristics of hyperspectral data.

Images