摘要
高光谱遥感影像智能解译是实现高光谱遥感应用的重要研究任务之一。针对生成式对抗网络在高光谱遥感影像分类中空谱特征利用不足的问题,提出了一种基于CVAE-GAN的高光谱遥感影像分类对抗网络算法(hyperspectral remote sensing classification based on CVAE-CGAN,HCVAE-CGAN),通过搭建1D-CNN分类模型和2D-CNN分类模型,训练判别器识别空谱特征,利用CVAE替代生成器结构生成影像光谱特征和空间特征,通过encode模块处理训练集得到空谱特征值,并将空谱特征值解码生成图像光谱,随后比对原始图像进行decode网络模型的优化,最后利用生成样本对分类器进行训练。实验结果表明,HCVAE-CGAN方法在小样本训练中有更好的检测性能,在Indian Pines和Pavia University数据集中的总体精度分别提高了2.85个百分点和3.92个百分点。
Intelligent interpretation of hyperspectral remote sensing images is one of the important research tasks in hyperspectral remote sensing applications.In order to solve the problem that the model of generative adversarial network is insufficient utilization of spatial and spectral features in hyperspectral remote sensing image classification,in this paper,a hyperspectral remote sensing classification based on CVAE-CGAN(HCVAE-CGAN)is proposed.By building 1D-CNN classification models and 2D-CNN classification models,the discriminator is trained to recognize empty spectral features,and the CVAE replacement generator structure is used to generate image spectral features and spatial features.The empty spectral characteristic values are obtained by processing training sets through encode module,and the empty spectral characteristic values are decoded to generate the image spectrum.Then,the original images are compared to decode network model optimization.Finally,the generated samples are used to train the classifier.The experimental results show that the HCVAE-CGAN method has better detection performance in small sample training,and the overall accuracy in Indian Pines and Pavia University data sets improves by 2.85 and 3.92 percentage points,respectively.
作者
郑猛猛
葛小三
ZHENG Mengmeng;GE Xiaosan(School of Surveying and Mapping and Land Information Engineering,Henan Polytechnic University,Jiaozuo,Henan 454003,China)
出处
《遥感信息》
CSCD
北大核心
2024年第1期83-92,共10页
Remote Sensing Information
基金
河南省自然科学基金(222300420450)
国家自然科学基金(41572341)。
关键词
高光谱图像分类
生成式对抗网络
分类方法
深度学习
hyperspectral image classification
generative adversarial network
classification method
deep learning
