基于生成式对抗网络的煤矿图像去噪研究  

Research on Coal Mine Image Denoising Based on Generative Adversarial Network

作　　者：孔华永 娄昭远 Kong Huayong;Lou Zhaoyuan(Guoneng Wangxin Technology(Beijing)Co.,Ltd.,Beijing 100000,China)

机构地区：[1]国能网信科技(北京)有限公司,北京100000

出　　处：《煤矿机械》2022年第7期173-175,共3页Coal Mine Machinery

摘　　要：为了去除煤矿图像的噪声并尽可能地还原煤矿图像的原貌,提出了一种基于生成式对抗网络(GAN)的煤矿图像去噪方法。首先考虑到模型的输入为含噪声的煤矿图像,因此利用变分自编码器(VAE)作为GAN的生成器,以获得不含噪声的煤矿图像。然后考虑到VAE生成的重构图像一般情况下均是较为模糊的,因此利用GAN中的判别器对原始图像与重构图像进行判断,以提升去噪图像的清晰度。最后将作为生成器的VAE与GAN中的判别器进行组合,设计了一种适用于煤矿图像去噪的深度学习算法。实验表明,该方法在各精度指标上都有很好的表现,且去噪图像均较为清晰。In order to remove the noise of coal mine images and restore the original appearance of coal mine images as much as possible, a coal mine image denoising method based on generative adversarial network(GAN) was proposed. First of also, considering that the input of the model is a noisy coal mine image, a variational auto-encoder(VAE) was used as the generator of the GAN to obtain a noise-free coal mine image. Then, considering that the reconstructed image generated by the VAE is generally fuzzy, the discriminator in the GAN was used to judge the original image and the reconstructed image to improve the clarity of the denoised image. Finally, combining the VAE as a generator with the discriminator in the GAN, a deep learning algorithm suitable for denoising coal mine images was designed. Experiments show that the method has a good performance in all accuracy indicators and the denoising images are relatively clear.

关 键 词：煤矿图像 图像去噪 深度学习 GAN VAE 

分 类 号：TD76[矿业工程—矿井通风与安全]