Research On Core Image Segmentation Algorithm Based On Deep Learning

Core image is one of the most important geological data in the process of oil and gas field development.It is of great significance to understand the underground geological condition,determine the core lithology and infer the sedimentary environment.With the development of deep learning,semantic segmentation of core image using deep learning method has become a research hotspot.In view of the problems existing in existing methods in core image segmentation,such as excessive consumption of computing resources,easy overfitting of models,gradient disappearing or explosion,pixel redundancy,etc.,on the basis of previous studies,a core image semantic segmentation algorithm based on deep learning is proposed.Specific research work is as follows:(1)Experimental core image data sets are produced.Firstly,the labelme annotation tool was used to segment the core image into rock,liquid in rock cracks and natural gas.Secondly,the core image data set is cut and flipped to expand the data set.Finally,the expanded core image and the segmentation image of the core image are divided into the corresponding training set,verification set and test set.(2)An HRNet core image segmentation algorithm based on multi-scale context information is implemented.Because of the parallel superposition of pyramid pool module network,the feature information of each branch cannot be shared.In this paper,the pyramidal pool module is improved into a network model based on jump connection structure,which makes certain connections between cavity convolution with different expansion rates and forms a denser network structure.Then,the improved pyramid pool module is introduced into each subnet of the HRNet algorithm,which alleviates the redundant phenomenon of the HRNet network model and realizes the task of core image segmentation.Compared with the current popular semantic segmentation algorithm,the improved algorithm achieves better segmentation effect by comparing the experimental results and analyzing the evaluation index value.(3)An EASPP-HRNet core image segmentation algorithm based on residual density is implemented.In this paper,Re LU,PReLU and FReLU activation functions are compared first,and then three activation functions are introduced into EASPP-HRNet core image segmentation algorithm respectively,which proves that FReLU activation function has better effect on image segmentation.Then,this paper introduces the ordinary residual network and the dense residual network into the EASPP-HRNet core image segmentation algorithm respectively,and proves that the dense residual network is better for image segmentation.The improved algorithm solves the problems such as excessive consumption of computing resources,easy overfitting and gradient disappearing or explosion.After comparison and analysis of experimental results,the improved algorithm has better segmentation effect and region segmentation effect.The comprehensive experimental results show that the improved HRNet semantic segmentation algorithm has a better effect,and its segmentation effect is closer to the actual core image segmentation effect,which is of great significance for further understanding of underground geological conditions,determining core lithology and inferring sedimentary environment.