利用卷积神经网络的高光谱图像分类

针对高光谱图像分类中对光谱信息利用不足的问题,提出一种基于卷积神经网络在光谱域开展的分类算法。该算法通过构建五层网络结构,逐像素对光谱信息开展分析,将全光谱段集合作为输入,利用神经网络展开代价函数值的计算,实现对光谱特征的提取与分类。实验中采用三组高光谱遥感影像数据进行对比分析,以India Pines数据集为例,提出的基于卷积神经网络的分类方法的分类正确率达到90.16%,比RBF-SVM方法高出2.56%,相比三种传统的深度学习方法高出1%~3%,训练速度也较为理想。实验结果表明,本文所提出的算法充分利用了高光谱图像中逐像素点的光谱域信息,能够有效提高分类正确率。与传统学习算法相比,在较少训练样本的情况下,更能发挥其良好的分类性能。

Convolutional Neural Networks for Hyperspectral Image Classification

In order to solve the problem of insufficient utilization of spectral information in high spectral image classification,convolutional neural networks(CNN) are employed to classify hyperspectral images directly in spectral domain.More specifically,the architecture of the proposed classifier contains five layers.The collection of all optical spectra is taken as input by pixel,and the value of the cost function is calculated to realize the extraction and classification of the spectrum feature.Three sets of hyperspectral remote sensing image data are used as the experimental objects.Taking Pines India data set as an example,the rate of correct classification of the hyperspectral image based on convolutional neural networks is 90.16％.Compared with the RBF-SVM method and the three kinds of traditional methods of deep learning,the classification accuracy of the CNN method is 2.56％ and 1％ ～ 3％ higher than those of the methods separately.And the training speed is the most ideal.The experimental results show that the proposed algorithm makes full use of pixel spectral information of the hyperspectral image and it can effectively improve the rate of correct classification.Compared with traditional learning algorithms,it can achieve better classification performance with fewer training samples.