基于多维度特征遥感图像分类方法的研究

Research on remote sensing image classification method based on multi-dimensional features

为了解决传统高光谱图像分类方法精度低、计算成本高及未能充分利用空-谱信息的问题,本文提出一种基于多维度并行卷积神经网络(multidimensional parallel convolutional neural network,3D-2D-1D PCNN)的高光谱图像分类方法。首先,该算法利用不同维度卷积神经网络(convolutional neural network,CNN)提取高光谱图像信息中的空-谱特征、空间特征及光谱特征;之后,采用相同并行卷积层将组合后的空-谱特征、空间特征及光谱特征进行特征融合;最后,通过线性分类器对高光谱图像信息进行精准分类。本文所提方法不仅可以提取高光谱图像中更深层次的空间特征和光谱特征信息,同时能够将光谱图像不同维度的特征进行融合,减小计算成本。在Indian Pines、Pavia Center和Pavia University数据集上对本文算法和4种传统算法进行对比实验,结果表明,本文算法均得到最优结果,分类精度分别达到了99.210%、99.755%和99.770%。

In order to solve the problems of low accuracy,high computational cost and failure to make full use of space spectrum information of traditional hyperspectral image classification methods,a hyperspectral image classification method based on multi-dimensional parallel convolution neural network(3 D-2 D-1 D PCNN)is proposed in this paper.Firstly,the algorithm uses different dimensions of convolutional neural network(CNN)to extract the spatial spectral features,spatial features and spectral features of hyperspectral image information.Then,the same parallel convolution layer is used to fuse the combined spatial spectral features,spatial features and spectral features.Finally,hyperspectral image information is accurately classified by linear classifier.The proposed method can not only extract the deeper spatial and spectral feature information in hyperspectral images,but also fuse the features of different dimensions of spectral images to reduce the computational cost.Comparative experiments are carried out on Indian Pines,Pavia Center and Pavia University data sets.The results show that the proposed algorithm obtains the optimal results,and the classification accuracy reaches 99.210%,99.755%and 99.770%respectively.