摘要
传统的高光谱图像分类模型只考虑光谱特征信息,忽略了图像空间结构信息在分类中的重要作用。为提高高光谱遥感图像的分类精度,提出一种同时利用高光谱图像的光谱信息和空间信息的深度卷积神经网络分类模型。通过对低层特征自动分层地学习来提取更加抽象的高层特征,提取的特征具有平移、缩放及其他形式扭曲等高度不变性;基于学习到的深度特征,用logistic回归分类器进行分类训练。高光谱数据实验结果表明,深度卷积神经网络模型能够提高高光谱遥感图像的分类精度,从而验证了深度卷积神经网络进行高光谱图像分类的可行性和有效性。
The traditional hyperspectral image classification model only considers the spectral feature information, and ignores the important role of image spatial structure information in classification. In order to improve the classification accuracy of hyperspectral re-mote sensing image, this paper present a deep learning model utilizing the rich spectral and spatial information in hyperspectral images for land cover classification application. The proposed model is able to automatically extract more abstract high-level features from the low-level features for classification. In addition, the network structure is highly invariant to translation, scaling and other forms of dis-tortion. Experiment results show that the deep learning method can provide high performances in hyperspectral image classification ap-plications. The feasibility and effectiveness of the deep convolution neural network for classification of hyperspectral images are verified.
出处
《西华大学学报（自然科学版）》
CAS
2017年第4期13-20,共8页
Journal of Xihua University:Natural Science Edition
基金
水资源与水电工程科学国家重点实验室开放基金资助项目(2014SWG04)
国土资源部地学空间信息技术重点实验室开放基金(KLGSIT201411)
广西空间信息与测绘重点实验室开放基金(140452413、GKN 120711516)
关键词
高光谱遥感图像
卷积神经网络
特征提取
logMc回归分类器
分类精度
可行性
有效性
hyperspectral remote sensing image
deep convolutional neural network
feature extraction
logistic regression classifi-er
classification accuracy
feasibility
effectiveness
