A hyperspectral band selection method and device based on band attention reconstruction network

Abstract

The invention relates to the field of hyperspectral remote sensing image processing, and discloses a hyperspectral band selection method and device based on a band attention reconstruction network. Including: (1) construct an efficient band attention reconstruction network, and calculate the band representativeness; (2) select the band with the highest representativeness measure value, and construct the initial selected band subset; (3) calculate the redundancy between the bands ; (4) Calculating the comprehensive score of band representativeness and redundancy; (5) Selecting the band with the highest score and updating the band subset. Repeat process (3)‑(5) until the number of selected bands reaches the preset number of bands. The present invention starts from the characteristics of hyperspectral images, utilizes the complex nonlinear relationship between the bands, combines advanced deep learning knowledge, and proposes a band selection method that takes into account both band representation and redundancy, which can improve the classification of hyperspectral image pixels accuracy.

Description

technical field [0001] The invention relates to the field of image processing, in particular to a hyperspectral band selection method and device based on a band attention reconstruction network. Background technique [0002] Hyperspectral images consist of hundreds of bands and contain rich spatial and spectral information, making it possible to accurately identify objects of interest. However, in practical applications, the data redundancy brought by a large number of bands leads to the "Hughes phenomenon", which also brings a heavy computational burden to the subsequent image processing. Therefore, a reasonable data dimensionality reduction method becomes the key to using hyperspectral images. Currently, dimensionality reduction methods for hyperspectral images can be divided into two categories: feature extraction and band selection. The former will make it difficult to interpret the physical meaning of the data after dimensionality reduction, while the latter can prese...

AI Technical Summary

Problems solved by technology

[0007] In view of the deficiencies in the prior art, the purpose of the present invention is to provide a hyperspectral band selection method and device based on an efficient band attention reconstruction network, revealing the complex nonlinear relationship between the hyperspectral image bands, and at the same time through the band selection The redundancy and representativeness of the bands are considered in the design of the strategy to improve the extraction effect of the best band subset of hyperspectral images, and solve the problem of low classification accuracy of hyperspectral images caused by a large amount of redundant information between bands

Images