一种生成对抗网络半监督遥感图像分类方法

A Semi-supervised Remote Sensing Image Classification Method on Generative Adversarial Network

遥感图像由于数据集小,有标签数据少,因此其分类精度往往不高。为了提高遥感图像的分类精度,结合生成对抗网络与VGGNet-16设计了一个针对遥感图像的半监督分类方法,并分别在NWPU-RESISC45数据集与UC-Merced数据集上进行了验证。实验结果表明,该方法不仅能生成大量质量较好的遥感图像,增广了遥感图像数据集,解决了原始数据集样本不足的问题,同时能充分利用这些数据达到提高分类精度的效果,缓解有监督分类需要用到大量有标签数据的问题。

The classification accuracy of remote sensing images is often not high because of the small data set and few labeled data.In order to improve the classification accuracy of remote sensing images,a semi-supervised classification method for remote sensing images is designed in combination with generative adversarial network and VGGNet-16,and verified on NWPU-RESISC45 dataset and UC-Merced dataset respectively.The experimental results show that the method proposed in this paper cannot only generate a large number of well-quality remote sensing images,expand the remote sensing image dataset,solve the problem of insufficient samples of the original dataset,but also make full use of the data to improve the accuracy of classification,and alleviate the problem of using a large number of labeled data for supervised classification.