Reconstruction Of CT Images Of Carbonate Reservoir Based On Generative Adversarial Network

Carbonate reservoirs occupy an important position among the discovered oil and gas reservoirs in the world.Most of the reservoirs with large reserves and high production are carbonate reservoirs.Due to the influence of structure,sedimentation and diagenesis,carbonate reservoirs have diverse pore structures,complex reservoir structures and strong heterogeneity.CT technology has fast,non-destructive,and three-dimensional characteristics.It is considered to be the most effective method to obtain the internal structure of the rock without damage.However,in the process of collecting CT images of carbonate rocks,due to the limitation of the detector,a trade-off between the image field of view and the image resolution is required.Under these restrictions,it is impossible to directly obtain ideal high-resolution images.Numerous researchers have paid extensive attention and in-depth research to solve this problem by reconstructing super-resolution images from low-resolution images.The traditional super-resolution reconstruction method not only has low algorithm efficiency,but also loses a lot of detailed information,which makes the reconstructed image quality poor.In recent years,the deep learning-based super-resolution reconstruction method has greatly improved the efficiency of the algorithm and the quality of the reconstructed image.However,there are still problems such as unstable network,inability to restore high-frequency information,lack of key detail textures,and poor visual effects.In view of these issues,based on the residual structure,attention mechanism and generative adversarial network,we deeply investigate the super-resolution reconstruction of core CT images,and propose a variety of deep learning methods to improve the quality and reconstruction efficiency of digital core reconstruction images.The presented algorithms in this study are summarized as follows:An improved super-resolution reconstruction algorithm based on the super-resolution residual network model(SRRes Net)is proposed.The traditional SRRes Net model algorithm has the problems of large parameter,high computational complexity,and low network flexibility.We modified the structure of SRRes Net,deleted part of the redundant batch normalization layer in the residual block,and replaced the original Re LU activation function with the ELU activation function,which accelerates the convergence of the network and improves the training speed of the network.In order to solve the problem of excessively smooth edges of the reconstructed super-resolution images,the VGG19 model is introduced to calculate the perceptual loss of the network,and the content loss is merged with the added perceptual loss to increase the texture information of the image.Experimental results show that the proposed algorithm not only has a great improvement in the pixel accuracy,but also has a great improvement in the detail texture,which enhances the quality of the reconstructed images.A super-resolution reconstruction algorithm based on residual channel attention mechanism(AM-ASRRes Net)is proposed to solve the problem of blurring of high-frequency information and lack of detailed information in the image.In the AM-ASRRes Net,the residual block is converted into the residual channel attention block in the SRRes Net network.Because the attention mechanism is directly embedded in the backbone network,it will greatly reduce the network's ability to extract features and affect the restoration of image details and textures.In order to achieve effective calculation of channel attention and infer finer channel attention,the channel attention module based on global average pooling is connected in parallel with the channel attention module based on global maximum pooling.The jump link is used to connect to the residual module,which enhances the network's ability of feature extraction.For solving problems such as unsatisfactory reconstruction of image color and brightness,and model complexity,all batch normalization layers in the network are deleted,which speeds up the training of the network and improves the quality of the reconstructed images.Experimental results show that the presented algorithm can better process the texture information and highfrequency information of the reconstructed image,enrich the detailed texture information,and obtain clearer and high-quality super-resolution images.A super-resolution reconstruction algorithm based on residual channel attention and generative adversarial network is proposed.It is named as the super-resolution dual-channel attention generation adversarial network(DCA-SRGAN),which solves the problem of smooth edges,loss of detail texture and unrealistic reconstruction of the images.In the DCA-SRGAN,the AM-ASRRes Net network based on residual channel attention is used as a generator of the generate adversarial network.In order to obtain more contextual information and reduce the consumption of computing resources,two channel attention modules based on global average pooling are added to the middle and rear parts of the discriminator.The experimental results show that the visual effects of the reconstructed images obtained by DCA-SRGAN are better than other algorithms.DCA-SRGAN can make the detailed texture information richer.In the enlarged details of the reconstructed images,the edges of the images can be seen to be clearer and sharper.The proposed method significantly improves the quality of the reconstructed images.