Classification of Hyperspectral Images Using Conventional Neural Networks

We show that for the classification of fragments of a hyperspectral image, it is very effective to first transform its spectral features into principal components and then to recognize it using a convolutional neural network trained on a sample composed of fragments of this image. High percentage of correct classification was obtained when working with a large-format hyperspectral image while some of the classes of the hyperspectral image are very close to each other and, accordingly, are difficult to distinguish by hyperspectra. We investigate the dependence of the correct classification on the change in the size of the fragments from which the training and validation samples are composed and on the parameters of the convolutional neural network.