Computer Engineering and Applications ›› 2020, Vol. 56 ›› Issue (21): 199-203.DOI: 10.3778/j.issn.1002-8331.1909-0249

Previous Articles     Next Articles

Research on Super-Resolution Reconstruction of Remote Sensing Image Based on Improved Conditional Generative Adversarial Networks

JIANG Wenjie, LUO Xiaoshu, DAI Qinxuan   

  1. College of Electronic Engineering, Guangxi Normal University, Guilin, Guangxi 541004, China
  • Online:2020-11-01 Published:2020-11-03

基于对抗网络遥感图像超分辨率重建研究

蒋文杰,罗晓曙,戴沁璇   

  1. 广西师范大学 电子工程学院,广西 桂林 541004

Abstract:

Image resolution is an important indicator to measure the quality of remote sensing images. It is limited by imaging equipment and transmission conditions. The resolution of traditional remote sensing images is difficult to guarantee. To solve the above problems, a super-resolution reconstruction of remote sensing image based on conditional generation confrontation network is proposed. In order to speed up the convergence of the model, the combination of content loss and anti-loss is used as the objective function in the generator network. In addition, in order to improve the stability of network training, a gradient penalty function is introduced in the discriminator network to limit the discriminator gradient. The experimental results show that the improved model has significant improvement in subjective visual effects and objective evaluation indicators compared with SRCNN, FSRCNN and SRGAN models.

Key words: remote sensing image, super-resolution, conditional generative adversarial networks, gradient penalty

摘要:

图像分辨率是衡量遥感图像质量的重要指标,受限于成像设备和传输条件,传统遥感图像的清晰度难以保证,针对上述问题,提出了一种基于条件生成对抗网络的遥感图像超分辨率重建的改进模型。为了加快模型的收敛速度,在生成器网络中使用内容损失和对抗损失相结合作为目标函数。另外为了提高了网络训练的稳定性,在判别器网络中引入梯度惩罚函数对判别器梯度进行限制。实验结果表明,改进后的模型相较于SRCNN、FSRCNN和SRGAN模型,主观视觉效果和客观评价指标均有显著提升。

关键词: 遥感图像, 超分辨率, 条件生成对抗网络, 梯度惩罚