基于生成对抗网络的高光谱图像分类

为了解决简单卷积神经网络(convolutional neural network,CNN)不能有效提取与充分利用高光谱图像特征信息的问题,提出了一种基于残差网络的多层特征匹配生成对抗网络模型.提出的模型引入残差网络以挖掘高光谱图像的深层特征,生成可分性更高的高光谱图像,并通过一个特征融合层进行特征融合,充分利用网络的...

Hyperspectral image classification based on generating adversarial network

In order to solve the problem that simple (convolutional neural network, CNN) cannot effectively extrac t and make full use of the feature information of hyperspectral images,a multi-layer feature matching generative adversarial network model based on residual network was proposed.In theproposed model,residual network is introduced to mine the de ep features of hyperspectralimages to generate hyperspectral images with higher separability,and feature fusion is performed through a feature fusion layer to make full use of the features of each layer of thenetwork.The proposed algorithm achieves a classification accuracy of 95.6%,99.2% and 991% on Indian Pines,Pa via University and Salinas datasets,respectively.Compared with radial basis function-support vector machine(RBF-SVM),stacked autoencoder(SAE),deep belief network(DBN),CNN based on pixel-pair feature (PPF-CNN),CNN and three-dimensional convolutional neural network(3D-CNN),the proposed algorithm achieves aclassification accu racy of 99.1% on Indian Pines,Pavia University and Salinas datasets.The classifi cation accuracy is improved obviously.Experimental results show that the propose d method is an effective method for hyperspectral image classification.