【摘要】 岩心图像是油气勘探、地质分析的宝贵资料，由于部分旧设备、低分辨率设备采集到的图像细节不够，或因压缩过程中造成的图像细节丢失，容易影响地质工作者的判断，从而影响工作进度。为了获得更加真实、重建视觉效果更好的岩心图像，提出了一种改进的岩心图像超分辨率重建算法Improving-SRGAN。该算法采用生成对抗网络框架，将原生成器模型中的残差模块替换为残差注意力RAM模块。RAM模块可以对输入的特征图进行空间域和通道域加权，以更好地提取图像的高层次语义信息。同时在判别器中嵌入空洞空间金字塔池化ASPP模块，提升网络对岩心图像多尺度信息的感知能力，从而更准确地评估生成器生成图像的质量。相比SRGAN算法，所提出的算法在自建岩心数据集上的实验数据表明，其PSNR和SSIM分别提高了1.469、0.074；在DIV2K＿Valid数据集上的PSNR和SSIM分别提高了0.721、0.017，且算法的MOS得分更高，重建视觉效果优于其他算法。更多还原

【Abstract】 Core images are valuable data for oil and gas exploration and geological analysis, and due to insufficient image details collected by some old equipment and low-resolution acquisition equipment, or the loss of image details caused by the compression process, it is easy to affect the judgment of geologists, thereby affecting the progress of work. In order to obtain more realistic and better reconstructed high-resolution core images, an improved core image super-resolution reconstruction algorithm Improving-SRGAN is proposed. The algorithm adopts the generative adversarial network framework, and replaces the residual attention module in the original generator model with the residual attention module(Residual Attention Module). The RAM module can weight the spatial domain and channel domain of the input feature map to better extract the high-level semantic information of the image. At the same time, the Atrous Spatial Pyramid Pooling ASPP module is embedded in the discriminator to improve the network′s perception of the multi-scale information of core images, so as to more accurately evaluate the quality of the images generated by the generator. Compared with the SRGAN algorithm, the experimental data of the proposed algorithm on the self-built core dataset show that the PSNR and SSIM are increased by 1.469 and 0.074, respectively, while the PSNR and SSIM on the DIV2K＿Valid dataset are increased by 0.721 and 0.017, respectively, and the MOS score of the algorithm is higher, and the reconstruction visual effect is better than that of other algorithms.更多还原