Theory And Application Of Deep Data Mining And Intelligent Analysis For Geological Rock Mass Images Of Hydraulic Engineering

The geological survey and analysis are crucial in hydraulic engineering.However,the information and guidance are not enough in a high slope or deep tunnel with complex geological conditions.It is time-consuming and laboursome to record effective geological data,which makes it hard to realize the real-time analysis and feedback.On the other side,some of the data,such as the geological images,are only saved as documents.The unstructured data cannot be applied in calculation and analysis.In recent years,with the rapid development of computer science,mathematics,and geological information science,some new technologies provide methods for geological data recording and analysis.Deep learning based on big data has been applied widely in many fields,which is effective in the unstructured data analysis,i.e.images.The application of deep learning in geological survey and analysis can improve the automation of geological data processing,which can reduce the workload of the geological engineers and increase the intelligence of geological data recording and analysis.In this research,the deep learning involved in geological rock mass analysis methods in hydraulic engineering is applied to analyze the real image data in hydraulic engineering.The image data include different scales of data,from large scale geological structures to small scale borehole images.A systematic analysis is made to explore geological image data analysis.The main achievements and research contents are as follows:(1)The geological structure image data in the geological survey is taken as the research object.Based on the extracted features using different methods,the hydraulic geological structure identification model can be established.Through the comparison of the machine learning algorithms,convolutional neural networks,and transfer learning based on the deep learning models using the extracted features,the optimized model can be selected.The accuracy is chosen to evaluate the model.The influence of color and texture on the result is also discussed.The final result shows that the optimal model can provide support for the geological survey in hydraulic engineering.(2)The multiple deep learning models are adopted to extract the features of the basic geological phenomenon in the tunnel.Based on the extracted features,several machine learning algorithms are adopted to build the optimal identification model.The deep learn models and machine learning algorithms are combined to build the outstanding classification models for the basic geological phenomenon in the tunnel.A comparison among the machine learning algorithms is made to search a suitable one.The optimal model can support the tunnel design in engineering.(3)Based on the borehole images,a comparison between the traditional image processing methods and the object detection methods based on deep learning has been made.Furthermore,we also compare the performance of all the object detection models using different deep learning models.The result shows that the traditional methods based on the pixel threshold and edge cannot detect the geological boundary correctly.Because the geological condition is complex and there are so many noises in the images.It is hard to recognize the correct object only depending on the pixel or edge features.While the models based on the deep learning models can ignore the noises and get the effective features of the borehole images.They can detect the right position of the geological boundaries.The convergence process of model training is fast and the error is small in the model training process.The effective detection of geological boundaries is significant in improving the automation and intelligence in the 3D geological model establishment.(4)The fractures in borehole images are taken as the research object.The traditional image segmentation methods and the deep learning models are compared in the application of borehole fractures segmentation.Transfer learning is adopted based on different pre-trained models.The result of the comparison can show the performance of different models.The model with the best performance is selected as the optimal model for borehole image fracture segmentation.The traditional methods are fast but with low accuracy.The segmentation models based on deep learning are also fast in image testing after training.The segmentation model can recognize the object at a pixel level with high accuracy.The multiple segmentation models based on deep learning are trained using transfer learning.The convergence process is also discussed.Through the comparison of the traditional methods and multi segmentation models based on deep learning,the borehole image fracture analysis method is proposed,which is meaningful in hydraulic engineering evaluation.