Hyperspectral image classification method based on automatically designed long-short term memory network

Abstract

The invention discloses a hyperspectral image classification method based on an automatically designed long-short term memory network. The method comprises the following steps: constructing a search space; building a controller by using a recurrent neural network; generating a training set and a verification set; generating a loop unit by the search space; training a long-short-term memory network built by a circulating unit; training the controller; and iteratively updating parameters of the controller, judging whether a strategy loss function is converged or not, if so, generating a circulating unit B from the search space by using the controller, training a long-short-term memory network NB established by the circulating unit B, and classifying the hyperspectral image to be classified, otherwise, continuing to generate the circulating unit by using the search space. The automatically designed long-short-term memory network has the advantage of making full use of the spectral information of the hyperspectral image to make the classification effect more accurate, so that the method can be used for classification of the hyperspectral image.

Description

Technical field [0001] The present invention belongs to the field of image processing, and further relates to a high spectral image classification method based on automatic design length when a high spectrum image classification technology. The present invention can be used to classify the object objects in high-spectral images, thereby providing a basis for resource exploration, forest coverage, and disaster monitoring field. Background technique [0002] In recent years, the application field of high-spectral images is increasingly wide, it has important value, can be used to disaster monitoring, resource survey, etc., and aspects of agriculture, geology and military. The high-spectral image is all pixel points to use hundreds of high-resolution continuous electromagnetic spectrum, so each pixel point contains rich spectral information. Rich spectral information makes high-spectral images show great advantages on classification tasks. In the past two decades, many traditional m...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to address the deficiencies of the above-mentioned prior art, and propose a hyperspectral image classification method based on an automatically designed long-short-term memory network, which is used to solve the problem of the need for manual design of the recurrent units in the long-short-term memory network. Knowledge, network model fixes the problem that it cannot be self-adapted to different hyperspectral images