Research On Algorithm For Identifying Pore Types In Shale Based On Deep Learning

In recent years,the extraction of shale oil has been a hot topic.As a substitute resource for oil,successful extraction of shale oil can alleviate the problem of insufficient oil reserves.China has abundant shale oil reserves,mainly in shale rock formations in some basin areas.The composition of rock components in shale formations is relatively complex,and rock pores are an important part of rock components.Identifying pore types is of great significance for reservoir quality evaluation and the formulation and implementation of shale oil extraction plans.Traditional pore type recognition has problems such as low recognition efficiency and accuracy,and strong subjectivity due to insufficient technical depth.In addition,small sample datasets are also the reason for the low efficiency and accuracy of pore type recognition.Deep learning has deep neural networks and the ability to automatically extract features from data,making it an excellent performer in the field of image recognition.In response to the above problems,this article improves the deep learning network model to improve the efficiency and accuracy of rock pore type recognition.The following research has been mainly conducted.To solve the problem of low recognition efficiency and accuracy caused by small sample data sets,ResNet18 is combined with transfer learning and fine-tuning to build a model,and the feature extraction ability of the model is improved by optimizing weight.Firstly,the data is expanded using image enhancement methods;Secondly,transfer the parameters on the big dataset ImageNet that do not include the classification layer to this article,and retrain the classification layer and ResNet18 top-level parameters for fine-tuning.Finally,the model is compared with the original model.The experiment shows that transfer learning and finetuning have greatly improved the performance and accuracy of the model due to the reduction of a large number of parameters.In order to further improve the accuracy of the model,the ResNet18 network model has been improved.Firstly,parallel asymmetric hollow convolution is used to replace ordinary convolution in residual blocks,improving the model’s feature representation ability;Secondly,in order to better capture the focus of feature maps,an attention mechanism module is added after the convolutional layer of the model,and the impact of attention mechanism on recognition performance at different positions of the model is experimentally compared.The final experiment shows that the improvement of ResNet18 model by parallel asymmetric hole convolution and attention mechanism is feasible,and the combination of the two has the best recognition effect.Finally,in order to efficiently and conveniently identify the types of shale pores,a platform for identifying shale pore types was developed based on the above research.Integrating the algorithm for identifying pore types studied in this article into the system,identifying the uploaded data,outputting pore types,and maintaining the algorithm,updating the algorithm for identifying pore types,improving the accuracy of recognition,is of great significance for the subsequent evaluation of reservoir quality.