Hyperspectral anomaly detection method based on full convolution cascade auto-encoder

Abstract

The invention discloses a hyperspectral anomaly detection method based on a full convolution cascade auto-encoder. The method comprises the following steps: 1) conducting encoding through a first encoder; (2) conducting decoding through a first decoder; 3) conducting encoding through a second encoder, wherein the second encoder encodes a first reconstructed image into a second potential feature; (4) conducting decoding through a second decoder, wherein the second decoder decodes the second potential feature into a second reconstructed image; and 5) determining an abnormal value of each pixel on the second reconstructed image by using the mahalanobis distance, obtaining second reconstructed data after network convergence, stretching the second reconstructed data into a two-dimensional matrix, performing anomaly detection by using the mahalanobis distance, and finally obtaining a 2D detection result image. According to the method, hyperspectral abnormal target detection is carried out through a full convolution cascade self-encoding network, and the defect that a single self-encoder is non-convex is overcome, so that the network can better find a globally optimal solution.

Description

Technical field [0001] The present invention belongs to the technical field hyperspectral image processing, abnormality detection relates to a method based on the whole hyperspectral concatenated convolutional encoder from. Background technique [0002] With the rapid development of space technology, remote sensing of the earth has become an important form of human observation of the Earth. Hyperspectral imaging technology combines imaging and spectroscopy, hyperspectral imaging is acquired a three-dimensional data cube. Viewed from the spatial domain can be seen as a series of stacked two-dimensional image; from the analysis spectral dimension, can be regarded as a feature and the reflected radiation in different wavelength bands. Abundant spectral information to target recognition and detection, scene classification, semantic segmentation provide strong support, the hyperspectral image processing and analysis is widely used in industry, agriculture, industry and other fields. ...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to provide a hyperspectral anomaly detection method based on a fully convolutional cascaded autoencoder, which solves the problems that the hyperspectral anomaly detection algorithm in the prior art is difficult to effectively use the characteristics of the middle layer, and the algorithm is cumbersome and complex.

Images