摘要
为解决标记样本缺乏、提升分类精度及增强模型容错性等问题,提出一种基于辅助分类器生成对抗网络(ACGAN)的分类方法.首先,将预训练的ACGAN模型作为光谱特征提取器,采用局部二值模式(LBP)算法提取图像的纹理特征;然后,融合光谱特征和纹理特征,由卷积神经网络(CNN)进行分类.在2个广泛使用的数据集上进行实验,结果表明:相较于其他方法,文中方法可显著提高分类精度.
In oder to solve the problem for lack of labeled samples,improve the classification accuracy and enhance the fault tolerance of the model,a hyperspectral sensing image classification method based on auxiliary classifier generative adversarial network(ACGAN)is proposed.Firstly,the pre-trained ACGAN model is treated as a spectral feature extractor,and the texture features of the image are extracted by local binary pattern(LBP)algorithm.Then,the spectral features and texture features are merged and calssified by convolutional neural network(CNN).Experiments on two widely used datasets show that compared with other methods,the proposed method can significantly improve the classification accuracy.
作者
刘群
陈锻生
LIU Qun;CHEN Duansheng(College of Computer Science and Technology,Huaqiao University,Xiamen 361021,China)
出处
《华侨大学学报（自然科学版）》
CAS
北大核心
2019年第1期113-120,共8页
Journal of Huaqiao University(Natural Science)
基金
国家自然科学基金面上资助项目(61370006)
福建省科技计划重点资助项目(2015H0025)
关键词
高光谱图像分类
生成对抗网络
局部二值模式
卷积神经网络
hyperspectral image classification
generative adversarial networks
local binary pattern
convolutional neural networks
