基于卷积神经网络的岩石图像分类

发布时间：2018-05-18 00:38

本文选题：岩石图像 + 图像分类　；参考：《西安石油大学学报(自然科学版)》2017年04期

【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。
[Abstract]:When analyzing and classifying rock slices under microscope, the artificial identification efficiency is low and easy to be affected by subjective factors. Therefore, an automatic classification method of rock granularity based on convolution neural network depth learning is proposed. In this method, the image feature is automatically extracted by convolution network model, and a pattern classifier is established at the same time to realize automatic granularity recognition based on thin slice image. The convolution network model was trained with 4 800 samples in Ordos Basin. The accuracy of the classification results of the test set was 98.555 through 1200 samples. Theoretical analysis and data validation show that the convolution network model established by depth learning can obtain efficient, accurate and reliable automatic classification results based on rock slice images.
【作者单位】：西安石油大学计算机学院;
【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)
【分类号】：TP183;TP391.41

【参考文献】