Hyperspectral Image Classification Based On Auto Encoder

The classification of hyperspectral images is an important element of hyperspectral remote sensing technology,which has important applications in both military and civilian fields.How-ever,due to the high dimensional features of hyperspectral images,high correlation between bands and other characteristics,the classification task faces great challenges.How to reduce the hyperspectral dimension and extract effective features is a special problem that needs to be considered by classification models.This paper investigates the hyperspectral image classification problem based on Auto En-coder model.The traditional depp Auto Encoder model is used to classify hyperspectral images,and there are problems such as the number of parameters is too large due to the full connection,and the training process is tedious layer by layer.In this paper,a new convolutional self-coding model is proposed for extracting the spectral spatial association of images.The model models the complex spatial structure of the images through the convolutional operations,which reduces the number of invalid connections and keeps the complexity of the model under control due to the local connectivity and parameter sharing features of the convolutional operations.In order to make better use of the information provided by hyperspectral image data,this paper proposes a semi-supervised classification model based on Auto Encoder model,which combines unsupervised and supervised learning to take full advantage of the information con-tained in the unlabeled data.The training of the model is divided into two phases: in the unsu-pervised learning phase,the model is trained with the whole data set? in the supervised learning phase,the model is further trained with the labeled data.In this paper,two hyperspectral images are classified separately,and the validity of the model is verified through the analysis of the experimental results.