Research On Intelligent Evaluation Method Of Hydraulic Rock Mass Quality Based On Depth Feature Of Core Image

In hydraulic engineering,rock mass quality evaluation and integrity analysis are important contents of hydraulic geological survey,and the evaluation results are important basis for the design and construction of hydraulic engineering.The integrity evaluation of rock mass is mainly completed by manual measurement and calculation of integrity evaluation ensemble.China’s large and medium-sized hydraulic engineering are mostly in deep valleys,and the geological exploration conditions are complex.It is time-consuming and laborious to use manual measurement.Therefore,establishing an accurate,efficient and automated rock mass quality evaluation method is an important content to be studied in the construction of hydraulic engineering.With the development of computer technology,image recognition technology has been widely used,especially the application of deep learning model,which greatly improves the recognition accuracy and provides technical support for intelligent evaluation of rock mass quality.Therefore,in this study,the borehole core image of water conservancy project is taken as the research object,and the core image data set is established through the preprocessing process of image collection,screening,image correction and annotation.A variety of deep learning algorithms are used to establish the intelligent quantitative model of core image classification and rock integrity designation.Taking the actual engineering core image as the test sample,the validity of the model and the integrity evaluation method is veri fied by experiments.The main research content is as follows:(1)An integrity evaluation model of hydraulic rock mass based on deep ensemble features of core image is established.The core images in practical engineering are collected,and the perspective transformation algorithm is used to remove the image clutter background and solve the problem of shooting angle deviation.Three deep learning models,Incption-v3,Inception-v4 and Inception-ResNet-v2,are applied to extract high-dimensional features of core images,and feature ensemble is realized by feature vector splicing.By comparing the applicability of different machine learning models to ensemble features,a rock mass quality classification model based on weighted support vector machine(WSVM)is established to realize the intelligent evaluation of hydraulic rock mass integrity.(2)A core geometric feature and RQD intelligent quantization method based on ENUNet model is proposed.Using ResNext,SEResNext,EfficientNet-b5,etc.as pre-training models,UNet as a framework model,and a ensemble segmentation model ENUNet is established through a weight design strategy.The ENUNet model is further used to realize the rapid identification and accurate segmentation of the core in the image.Finally,the length of the complete core is determined by contour detection,pixel statistics and waveform analysis,and the rapid quantitative calculation of the core length and RQD.(3)Establish an enchanced rock mass integrity evaluation designation ERQD.On the basis of summarizing the advantages and disadvantages of previous rock mass integrity evaluation methods,combined with the intelligent identification results of core length and type,an enchanced rock mass integrity evaluation designation ERQD is established.The designation introduces the function of core length,uses multiple linear regression length weight,and controls the influence of length on integrity designation by weight.At the same time,the influence of different core types such as complete core and broken core on integrity designation is considered.The application of actual engineering data shows that ERQD has better distribution law than RQD,and can more reasonably characterize the core characteristics.At the same time,the designation can also realize automatic calculation,which has application value in water conservancy project construction.