基于卷积神经网络的岩石图像分类  

Study on Rock Image Classification Based on Convolution Neural Network

作　　者：程国建[1] 郭文惠[1] 范鹏召 

机构地区：[1]西安石油大学计算机学院,陕西西安710065

出　　处：《西安石油大学学报（自然科学版）》2017年第4期116-122,共7页Journal of Xi’an Shiyou University（Natural Science Edition）

基　　金：国家科技重大专项(编号:2011ZX05044);陕西省工业科技攻关项目(编号:2015GY104)

摘　　要：在显微镜下分析岩石薄片并对其进行分类时,人工鉴定效率较低且易受主观因素影响,为此提出了一种基于卷积神经网络深度学习的岩石粒度自动分类方法。该方法通过卷积网络模型实现图像特征自动提取,并同时建立模式分类器,实现基于薄片图像的粒度自动识别。采用鄂尔多斯盆地的4 800样品对卷积网络模型进行训练,通过1 200个样品对模型测试,测试集分类结果的准确度达到98.5%。理论分析与数据验证说明,通过深度学习所建立的卷积网络模型能够基于岩石薄片图像获得高效、准确、可靠的自动分类结果。In the analysis of rock thin slices under microscope,the efficiency of artificial identification is low and the identification re- suh is easy to be influenced by subjective factors. In order to solve this problem, an automatic classification method of rock particle size based on deep learning of convolution neural network is proposed. The method realizes the automatic extraction of image features by means of convolution network model, and the automatic recognition of rock particle size based on slice images is realized by establishing pattern classifier. The convolution network model was trained by 4 800 samples from Erdos Basin, and the model was tested by 1 200 samples. The classification accuracy of the test set reached to 98.5%. Theoretical analysis and data verification show that the convolu- tion network model established by deeply learning can obtain efficient, accurate and reliable automatic classification results of rock parti- cle size based on rock slice images.

关 键 词：岩石图像 图像分类 岩石颗粒 卷积神经网络 

分 类 号：TE19[石油与天然气工程—油气勘探]