LiDAR Data Classification Based On Generative Adversarial Network

With the rapid development of Li DAR remote sensing technology,Li DAR image classification is an important branch in the field of remote sensing image processing.With the continuous development of image acquisition technology in recent years,more and more image classification methods can efficiently and accurately perform accurate classification.However,with the rapid development of deep learning,the use of deep model training requires a large number of training samples.However,in the field of remote sensing images,there are still problems such as too small number of training set samples.The training of the classifier needs to have more image samples to improve the classification accuracy.Based on generative adversarial networks,this paper studies the classification of Li DAR images and generates Li DAR images through GAN,which finally further improves the classification accuracy.In view of the shortcomings of this problem,the paper proposes two Li DAR image classification methods based on generative adversarial networks.First,several typical Li DAR classification methods,typical models for generating adversarial networks,and related basic knowledge used in this paper are introduced.Finally,classification evaluation indicators are introduced.Secondly,in view of the limitation of the number of data sets in the field of Li DAR image classification,a method of applying a generative adversarial network to Li DAR image classification is proposed.In this network model,the convolutional generation adversarial network is used to generate data that simulates real samples.The images and the original training samples are mixed into the training classification model,which improves the classification accuracy of Li DAR data.Finally,in order to improve the quality of the generated images,Caps Net is introduced as a powerful system reformer of CNN,which can represent the images more uniformly,and it is more robust to the changes in the pose and spatial relationship of some objects in the generated images.The dissertation proposes a Capsule GAN network that adds capsules to the GAN framework.The capsule network is used as a discriminator in the network model framework.The experimental results show that the classification accuracy of the trained classification model is better than that of CNN as a discriminator.