基于像素及梯度域双层深度卷积神经网络的页岩图像超分辨率重建

实际采集的页岩图像存在分辨率低等不足,有时难以满足实际应用的需求。针对此问题,构建了一种基于双层深度卷积神经网络的页岩图像超分辨率重建算法。算法以深度卷积神经网络为基础,引入残差训练及批规范化层来加速网络的收敛,并且在此神经网络的基础上提出图像像素域及梯度域结合的页岩图像超分辨率重建算法。算法大致过程为首先利用像素域的卷积神经网络对输入的低分辨率页岩图像进行上采样;然后对上采样图像提取梯度信息并利用梯度域的卷积神经网络对其进行转换;最后利用转换后的梯度信息作为正则项来约束高分辨率图像的重建,从而得到重建的高分辨率页岩图像。实验表明,与主流的超分辨率重建算法相比,重建得到的页岩图像具有更好的主观视觉效果与更高的客观评价参数,更利于后续的处理及分析。

Super-resolution of shale image using double deep convolutional neural networks in pixel-gradient domain

There are some problems in the actual shale image, such as low resolution, sometimes it is difficult to meet the needs of practical applications. To tackle with this problem, a super-resolution algorithm for shale image is proposed in this paper, which is based on double deep convolutional neural networks. In this algorithm, firstly, the input image is up-sampled by the pixel domain convolution neural network; Secondly, the gradient profile information is extracted from the up-sampled image and converted by the gradient domain convolution neural network. Finally, we use the converted gradient information as a constraint to reconstruct the high-resolution image. Moreover, we introduce the Batch-Normalization and deep residual-learning to improve the training speed of the neural network. The experimental results show that compared with the some leading super-resolution algorithm, the reconstructed image has a significant improvement in subjective vision and objective evaluation, and then it is helpful for the further processing and analysis of shale image.