Application Of Deep Transfer Learning In Hyperspectral Image Classification

Spectral features are widely used in hyperspectral image classification,but the inadequate separability of spectral features leads to low classification accuracy.At the same time,due to the different feature distribution of the same sample in different hyperspectral images,traditional methods can not effectively classify the same sample in multiple hyperspectral images.In this paper,residual network is used to extract the deep features of hyperspectral images,and domain adaptation network is used to align the feature spaces of different hyperspectral images,which improves the classification performance of hyperspectral images.The main work and results of this paper are as follows:1.Aiming at the problem of poor separability of spectral features in hyperspectral image classification.A hyperspectral image classification method based on residual network is proposed in this paper.Spatial feature of each pixel of hyperspectral image was extracted by using this method,the classification model is established based on residual network algorithm,and the gradient disappearance problem is avoided by using residual network structure,which improves the classification effect.2.Aiming at the problem of insufficient labeled samples in hyperspectral image classification,A method based on deep transfer learning theory is proposed that is to transfer some middle-level parameters of deep convolution network strategically,which realizes the transfer learning between two data sets of hyperspectral image,which improves the classification effect in the case of a small number of labeled samples.3.Aiming at the problem of feature space migration of similar samples in different data sets,In this paper,the domain adaptation network algorithm is used to align the feature space of two sets of hyperspectral image data to improve the classification effect of hyperspectral image without labeled data.