摘要
本文设计了一种改进的卷积神经网络（convolutional neural network，CNN）模型用于高光谱影像分类，该模型能够直接将高光谱影像数据立方体作为输入特征，不需要预先降维处理，且能够综合利用光谱和空间特征进行分类处理。实验结果表明，基于改进的CNN模型的高光谱影像分类方法比传统SVM、1D—CNN和PCA＋CNN等方法的分类精度更高。
An improved convolutional neural network model is designed for hyperspectral image classification. The model can directly use hyperspectral image data cube as input feature, without dimensional reduction processing in advance. Besides, hyperspectral images can be classified by comprehensively using spectral and spatial features. Experiment resuhs show that the spatial-spectral classification method of hyperspectral image based on CNN can achieve higher classification accuracy than tradi- tional methods such as SVM, 1D-CNN, PCA ＋ CNN, etc.
出处
《测绘科学与工程》
2017年第6期74-78,共5页
Geomatics Science and Engineering
关键词
高光谱影像
分类
卷积神经网络
空谱特征
hyperspectral image
classification
eonvolutional neural network
spatial-spectral features
