Publishing House SB RAS:

Publishing House SB RAS:

Address of the Publishing House SB RAS:
Morskoy pr. 2, 630090 Novosibirsk, Russia



Advanced Search

Avtometriya

2021 year, number 2

CLASSIFICATION OF HYPERSPECTRAL IMAGES USING CONVOLUTIONAL NEURAL NETWORKS

V.I. Kozik, E.S. Nezhevenko
Institute of Automation and Electrometry, Siberian Branch, Russian Academy of Sciences, Novosibirsk, Russia
Keywords: classification, hyperspectral image, principal components, convolutional neural networks, training

Abstract

For the classification of fragments of a hyperspectral image, a preliminary transformation of its spectral features to the principal components and subsequent recognition using a convolutional neural network trained on a sample composed of fragments of this image are demonstrated to be very effective. High percentages of the correct classification are obtained with a large-format hyperspectral image, despite the fact that some of the classes into which the hyperspectral image is divided are very close to each other and, accordingly, are difficult to distinguish by hyperspectra. The dependences of the correct classification on the size of the fragments, from which the training and validation samples are composed, and on the parameters of the convolutional neural network are investigated.