Super Resolution Reconstruction Of Core Image Based On Improved SRGAN

Digital core technology is a technology that scans the core through CT or electron microscope and converts it into digital data to simulate the real core.In recent years,it has been rapidly developed and widely used in petroleum,mining and other fields.However,due to technical reasons,the resolution of the digital core image is inversely proportional to the core size,which makes the resolution of the large-size sample image obtained by CT scanning low,while the sub sample resolution is high,but it is not universal,which is not conducive to the subsequent research on rock porosity,permeability,rock skeleton modeling,etc.Superresolution reconstruction technology is a technology to enhance the details of the image and eliminate the image blur caused by noise,compression,acquisition technology and so on by reconstructing the mapping relationship between low-resolution image and high-resolution image.Aiming at the task of super-resolution reconstruction of single image of core slice,this paper proposes an improved countermeasure generation network,which improves the residual module of the generator,reduces the number of network layers,and improves the running speed of the network.In addition,according to the new generator and the specific task of superresolution reconstruction of core image,a discrimination network with channel attention mechanism is designed,which makes the new discrimination network ensure that the texture details of the generated image have better expressiveness.The improved countermeasure network has excellent performance in the reconstruction task of core image,and the running speed is increased by 28% on the premise that the structural similarity is increased by 3%.In the task of super-resolution reconstruction of a single core image,the low resolution image of the core in the training data pair is obtained after the resolution of the high-resolution image is reduced by interpolation,but this simple degradation model can not reflect the degradation process of the real scene image.In order to further improve the reliability of the model,the large-scale sample image and sub sample image are registered by image registration technology to obtain the real low resolution image corresponding to the sub sample high resolution image.The image pair formed is used as the input of the model to realize the superresolution reconstruction of digital core image based on the reference image.The experimental test shows the effectiveness of the method.