基于SinGAN的岩石薄片图像超分辨率重建

Super-resolution Reconstruction of Rock Slice Image Based on SinGAN

岩石薄片图像对研究石油地质特性以及油气勘探都有重要的意义。由于各种因素的限制,获取到的岩石薄片图像经常会出现分辨率较低的情况,一定程度上限制了研究者对其细节信息的掌握。而一般的神经网络超分辨率算法都需要大量的数据作为训练集,为了提升岩石薄片图像超分辨率重建算法纹理细节信息还原能力,本文利用单图像生成式对抗网络,不需输入大量数据集,对岩石薄片图像进行超分辨率重建。采用鄂尔多斯某油区岩石铸体薄片图像进行训练,通过峰值信噪比(Peak Signal to Noise Ratio,SSNR)和结构相似性(Structural Similarity,SSIM)评价指标进行模型评价,实验结果表明:该方法超分辨率处理的图像在视觉效果和评价指标上均具有良好的效果。

The rock slice images are of great significance to the study of petroleum geological characteristics and the exploration of oil and gas.Due to the limitations of various factors,the obtained images of rock slices often have low resolution,which limits the researchers'grasp of their detailed information to some extent.General super-resolution algorithm of neural network will need a large amount of data as a training set,and in order to improve the ability of rock slice image super-resolution reconstruction algorithm to the restoration of the texture detail information of rock slice images,in this paper,the super-resolution reconstruction of single rock slice image is finished using single image generative adversarial network without inputting a large number of data sets.Rock cast thin section images from an oilfield area in Ordos were processed using this method,and the processing results are evaluated by the indexes of peak signal-to-noise ratio(PSNR)and structural similarity(SSIM)evaluation.It is shown that the processed images using this method has good visual effects and evaluation indexes.