Hyperspectral Image Classification Based On Generative Models And Deep Networks

Remote sensing detection technology has developed rapidly in recent years.There are more and more approaches to acquire remote sensing images,and the resolution is getting higher and higher.Hyperspectral image is an important part of remote sensing image,which contains both spectral information and spatial information of detection targets.Hyperspectral image is widely used in geological exploration,agricultural science and technology,city planning and other aspects.Many researchers have proposed various solutions for the processing and analysis of hyperspectral image,but most of the traditional methods cannot make full use of unlabeled data or cannot process redundant features.Classification and recognition technology for hyperspectral image still needs to be explored.With the development of high performance computing technology,deep learning methods have gradually shown better performance in the field of image processing.Faced with tasks such as image classification,detection,and generation,diverse artificial neural networks have been proposed successively.Under the conditions of analyzing the characteristics of hyperspectral data,this paper applies deep learning methods to the classification of hyperspectral image.The main contents are as follows:1)Aiming at the problem that traditional methods cannot use unlabeled data information,a hyperspectral image classification method based on evolutionary generative adversarial network is proposed.First,the powerful unsupervised learning ability of generative adversarial network is used to enable the method to extract the characteristics of unlabeled data;then The evolutionary algorithm is used to optimize the training process of generative adversarial networks,which uses multiple objective functions producing different generators to avoid the situation of pattern collapse and gradient disappearance.2)To solve the problem of the poor diversity of progeny and the possibility of learning degradation in the evolutionary generative adversarial network,a hyperspectral image classification method based on immune evolutionary generative adversarial network is proposed.Adopt a new generation method of borrowing from the idea of crossover,so that the descendant generators can be combined freely to the maximum extent,and the breadth of the search can be increased.The ideas of immune regulation and memory cells in the immune algorithm are borrowed to balance unsupervised and supervised learning Weights and prevent degradation.3)Aiming at the problem that the convolutional neural network model has many parameters and the computational overhead of processing redundant features is large,a hyperspectral image classification method based on EMP features and Ghost module is proposed.Method first uses the principal component analysis to reduce the dimensionality of the data,and then gets the extended morphological profile features of the dimensionality reduced data.Then the features were sent to a neural network model constructed by the Ghost module for classification.The Ghost module can generate redundant parts in the feature map with a small computational cost.Therefor it can reduce the computational cost of the model,and improve the operating efficiency without damaging the feature richness.