基于多尺度卷积神经网络的高光谱图像分类  

Hyperspectral Image Classification Based on Multi-Scale Convolution Neural Network

作　　者：刘娜[1] 宋娟[1] 董丽[1] 尤洪峰 LIU Na;SONG Juan;DONG Li;YOU Hongfeng(School of Intelligent Manufacturing,Qingdao Huanghai University,QingdaoShandong266427,China;School of Information Science and Engineering,Xinjiang University,UrumqiXinjiang830000,China)

机构地区：[1]青岛黄海学院智能制造学院,山东青岛266427 [2]新疆大学信息科学与工程学院,新疆乌鲁木齐830000

出　　处：《计算机仿真》2021年第7期451-455,466,共6页Computer Simulation

基　　金：山东省重点研发计划(2019GGX105001)。

摘　　要：卷积神经网络(CNNs)对高光谱图像(HSI)的特征具有强大的学习能力。但是由于传统卷积的本质,往往会丢失部分关键特征,给小样本分类增加了困难。为了克服上述问题,提出了一种新的多尺度卷积(MSC)运算方法来代替传统卷积运算。使用主成分分析(PCA)来降低高维的光谱特征并保留其关键特征;将MSC应用于几种经典的图像分类算法,并与它们进行了比较,实现了图像的分类。在三个公开的高光谱数据集上进行了实验,结果证明了提出的MSC是有效的。Convolutional neural networks(CNNs)have a powerful ability to learn the features of hyperspectral images(HSIs).However,due to the nature of traditional convolution,some key features are often lost,which makes it more difficult to classify small samples.In order to overcome this problem,in this paper,we propose a novel Multi-scale convolution(MSC)operation method to replace the traditional convolution operation.First,we used the principal components analysis(PCA)to reduce the hyperspectral data and retain its key features.Secondly,we applied MSC to several classic image classification algorithms and compared them.Experiments were carried out on three public hyperspectral data sets and the results show that the proposed MSC is effective.

关 键 词：多尺度卷积运算 小样本分类 高光谱影像 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]