Image Segmentation And Permeability Prediction Of Digital Rock Based On Deep Learning

China is rich in tight oil and gas reserves and has good prospects for exploration and development.However,the complex structure of tight rock and micro and nano-scale pores make it difficult to study the microscopic characteristics of reservoir rocks by traditional methods.Artificial intelligence technologies such as deep learning provide new methods for digital core processing and analysis,and new ways for predicting seepage patterns in tight rock reservoirs.In this paper,a set of digital core image processing and analysis methods for tight rocks based on artificial intelligence techniques is established,including automatic image segmentation,core permeability and seepage field prediction.Firstly,the core images are automatically recognized based on U-Net network,and evaluate the segmentation effect by accuracy and average intersection ratio,which not only eliminates the influence of human factors,but also improves the segmentation accuracy.The accuracy of binary segmentation is up to 99.87%,and the accuracy of multi-class segmentation is up to 96.77%.Secondly,combining the lattice Boltzmann and machine learning methods,the image features are automatically extracted to predict rock permeability.The analysis of the importance for image feature parameters shows that the hydraulic radius is closest to the rock permeability.Long short-term memory and random forest are successively used for permeability prediction,and compare the prediction performance with different input.Evaluate the model with RMSE,and the best results are obtained using random forest with all parameters input.Finally,this paper improves the U-Net model to predict the flow field of core images,and the average absolute error between the prediction results and the labels is 9.77×10^-5,and the computation time is saved by 98.59%.In summary,the automatic core image segmentation using U-Net network contributes to reduce human influence and improve work efficiency.The use of artificial intelligence technology to predict the permeability and the flow field is of great significance to the research of tight rock oil and gas development and fluid flow mechanism.