基于卷积神经ResNet50残差网络的岩石图像岩性识别研究
Lithology Recognition of Rock Image Based on Convolutional Neural ResNet50 Residual Network
王晓兵 1刘琳 1王俊卿 1胡石磊 1闻磊2
作者信息
- 1. 中核勘察设计研究有限公司,河南郑州 450000
- 2. 石家庄铁道大学工程力学系,河北石家庄 050043
- 折叠
摘要
深度学习卷积神经网络算法广泛应用于岩石图像的岩性识别.结合卷积神经残差ResNet50 网络,构建了岩石图像岩性识别模型,并依据定义的损失函数进行了网络模型的参数调优与验证;通过构建的识别模型对岩石图像岩性进行了预测,并根据识别结果进行了误差原因分析.研究表明:以深度卷积神经ResNet50 残差网络为基础,按照训练集、测试集、验证集 8∶1∶1 的比例可以进行岩性预测模型的构建与参数调优,从而实现岩石图像的岩性预测;结合项目实例构建了黑云母花岗闪长岩、变质砂岩、石英岩、黑云母花岗岩等四种岩性的岩石图像岩性识别模型;模型的识别准确率,除构造节理发育的破碎岩体较低外,一般可达 75%～90%;岩石图像识别结果的准确率受岩体构造裂隙发育及岩石图像质量影响较大,可以通过增加训练样本数量来提高识别结果准确率.
Abstract
Deep learning convolutional neural network algorithm is widely used in the lithology identification of rock images.A rock image lithology recognition model was constructed by combining the convolutional neural residual ResNet50 network,and the parameters of the network model were optimized and verified according to the defined loss function.At the same time,the li-thology of the rock image was predicted by the constructed recognition model,and the error causes were analyzed according to the pre-diction results.The research showed that based on the deep convolutional neural ResNet50 residual network,the lithology prediction model can be constructed and the parameters can be optimized according to the ratio of the training set,test set,and verification set 8:1:1,to realize the lithology prediction of rock image.Combined with the project example,the rock image lithology identification mod-el of four kinds of lithology,such as biotite granodiorite,metamorphic sandstone,quartzite,and biotite granite,is constructed.The re-cognition accuracy of the model is generally up to 75%～90%,except for the fractured rock mass with structural joints.The accuracy of rock image prediction results is greatly affected by the development of rock mass structural fissures and the quality of rock images.The accuracy of prediction results can be improved by increasing the number of training samples.
关键词
卷积神经网络/ResNet50/岩石图像/识别模型/岩性识别Key words
convolutional neural network/ResNet50/rock image/identification mode/lithology identification引用本文复制引用
基金项目
河北省自然科学基金重点项目(A2020210008)
出版年
2024
NETL
NSTL
万方数据