基于卷积神经网络的高光谱图像谱-空联合分类

随着深度学习的发展,卷积神经网络在各种视觉任务中都具有优越的性能;特别是在二维图像分类上,更是获得了很高的分类精度。针对于高光谱图像分类问题,设计了一种新的卷积运算;利用高光谱图像谱-空联合信息建立三维卷积神经网络对其进行分类;并针对高光谱图像样本不均匀性,在网络输出不同类别加入不同的权重加以训练。通过对两个公开高光谱图像数据集的测试,相对于传统方法,能够得到更高的分类精度,表明卷积神经网络对高光谱图像具有更强的特征表达能力。

Spectral and Spatial Classification of Hyperspectral Images Based on Convolutional Neural Networks

: With the development of deep learning, convolution neural network has superior performance in all kinds of visual tasks, especially the two-dimensional image classification in which it can get a high classification accuracy. In this paper, a new convolution operation was proposed in convolutional neural network according to the classification of hyperspectral imagery, and the 3d convolution neural network was established based on the spectral-spatial information in hyperspectral images. Aiming at the non-uniformity in samples of hyperspectral images, different weights in the different categories was added to the output network when training the network. Tested on the two open hyperspectral image datasets, this method can get higher classification accuracy than the traditional methods,which shows that convolutional neural network has the stronger expression of the characteristics in hyperspectral images.