基于卷积神经网络-深度迁移学习的岩性自动识别研究

Study on Automatic Recognition of Rock Lithology Based on Convolutional Neural Network and Deep Transfer Learning

基于人工智能算法实现岩性自动和快速识别是地质界和工程界的热点和难点,对于构建智能勘察体系具有重要意义。首先,基于地质学岩石分类体系建立系统的岩石图像数据集架构,用以支持后续岩石识别研究工作,该数据集包含约13000张岩石样本图像;其次,基于卷积神经网络-深度迁移学习算法,建立端到端、图像到标签的岩石图像智能识别模型,并对模型进行训练和测试;最后,模型泛化性和现场钻孔岩芯岩性验证试验表明,构建的岩石图像智能识别模型具有快速和准确识别岩石的能力,利用该模型对12种岩石的识别准确率达到95%以上。新提出的岩石图像识别模型可为现场地质和科研工作者提供方便和快捷的工具。

Automatic and fast rock classification identification based on artificial intelligence algorithm is a hot and difficult point in the geology and engineering fields,and it is of significance to build an intelligent survey system.Firstly,a systematic rock image dataset framework was established based on the geologic rock classification system to support subsequent rock identification studies.The dataset contains about 13 000 images of rock samples.Secondly,based on the convolutional neural network-depth transfer learning algorithm,an end-to-end,image-to-label intelligent recognition model for rock image classification model was established which was further trained and tested on dataset.Finally,the generalization and field drilling core validation tests were conducted,indicating that the rock image classification recognition model has an ability to quickly and accurately identify rock.On the basis of the model,an identification accuracy of over 95% for 12 types of rocks was obtained.The proposed rock image classification recognition model provides a convenient and fast tool for field geologists and scientific researchers.