Three-Dimensional Reconstruction Of Tight Sandstone Pores Based On CT Images And Kernel Clustering

Tight sandstone oil and gas,as the most important green energy in unconventional oil and gas resources,has great exploitation value.Among them,the pore structure of tight sandstone can provide effective information for the exploitation of oil fields.However,due to the complexity and diversity of pore structure,it is difficult to characterize it.Most of the existing researches are based on the pore recognition of two-dimensional sandstone images,and there are few studies on three-dimensional model characterization of pores based on scanning images of tight sandstone.Three-dimensional reconstruction method can be used to model the pores and show the internal structure and distribution of pores to oilfield workers.At present,the low segmentation accuracy of dense sandstone pores leads to a large gap between the three-dimensional reconstruction results and the actual sandstone pore structure.Therefore,image segmentation based on kernel clustering algorithm and three-dimensional reconstruction based on ray casting algorithm are used to construct the three-dimensional structure of pores,which more appropriately represents the three-dimensional structure of microscopic pores,a new idea for pore characterization is provided,and more valuable and referential information for the research on this basis also is presented.This thesis focuses on the three-dimensional reconstruction method of tight sandstone CT images based on kernel clustering,and superpixel semantic feature extraction based on the Efficient Net V2-S model,K-Means clustering algorithm base on the neural tangent kernel for superpixel merging,and three-dimensional reconstruction method base on ray casting algorithm are implemented.The main research focuses are as follows:1.Aiming at the problems caused by the K-Means clustering algorithm highly dependent on the sample distribution and using the mean method to update the clustering centers,Neural Tangent Kernel K-Means(NTKKM)clustering algorithm is proposed.Firstly,the Neural Tangent Kernel(NTK)is used to non-linearly map the input data to the high-dimensional space.Then,the K-Means algorithm combining the inter-cluster and intra-cluster distances is utilized to cluster the data in a high-dimensional space,and the clustering results are obtained.In order to prove the effectiveness of the method,experiments were conducted on public datasets,and the analysis of the experimental results show that the NTKKM clustering algorithm has better clustering results and higher stability compared with the traditional K-Means algorithm and the Gaussian kernel K-Means clustering algorithm.2.Aiming at the Pore segmentation of tight sandstone CT images,region merging algorithm based on kernel clustering(NTK-K-Means Clustering of Pore,NTK-KCo P)is proposed.Firstly,image filtering and image enhancement methods are used for preprocessing to highlight image features.Then,the SLIC0 algorithm is used to obtain the superpixels of the image.Next,using the characteristics of CT images of tight sandstone,the features of superpixels are extracted,which are semantic features,gray features and edge features.And then,the regions were merged by calculating the similarity of superpixel features and the similarity of superpixel positions.Finally,the pore segmentation effect of tight sandstone CT images are obtained.3.The three-dimensional reconstruction method of tight sandstone CT images based on ray casting algorithm is implemented.Firstly,the region merging algorithm based on kernel clustering is used to segment the pores of the dense sandstone CT image more accurately,which can reduce the calculation of invalid areas when using the ray casting algorithm.Then,trilinear interpolation and bounding box method are used to resaminate the sampling points.Finally,the front-to-back image synthesis method is used to render more realistic three-dimensionalrenderings of tight sandstone CT images.Finally,according to the above theory,a three-dimensional pore reconstruction system for tight sandstone CT images is established.