基于深度学习的智能矿物识别方法研究

doi:10.13745/j.esf.sf.2020.5.45

地学前缘 ›› 2020, Vol. 27 ›› Issue (5): 39-47.DOI: 10.13745/j.esf.sf.2020.5.45

基于深度学习的智能矿物识别方法研究

郭艳军¹^,⁴^,⁵(), 周哲², 林贺洵³, 刘小辉¹^,⁴^,⁵, 陈丹丘¹^,⁴^,⁵, 祝佳琪¹^,⁴^,⁵, 伍峻琦¹^,⁴^,⁵

1.北京大学地球与空间科学学院, 北京 100871
2.北京大学信息科学技术学院, 北京 100871
3.北京大学软件与微电子学院, 北京 102600
4.北京大学地球科学国家级实验教学示范中心, 北京 100871
5.北京大学地球科学国家级虚拟仿真实验教学中心, 北京 100871

收稿日期:2020-03-30 修回日期:2020-05-08 出版日期:2020-09-25 发布日期:2020-09-25
作者简介:郭艳军(1980—),女,博士,高级工程师,硕士生导师,主要从事信息地质、三维地质建模和地质大数据研究。E-mail: yanjunguo@pku.edu.cn
基金资助:
国家科技重大专项(2017ZX0513-002);教育部产学合作协同育人项目(201802267003);中央高校改善基本办学条件专项基金项目(XG2001221);北京大学本科教学改革项目(JG1901221)

The mineral intelligence identification method based on deep learning algorithms

GUO Yanjun¹^,⁴^,⁵(), ZHOU Zhe², LIN Hexun³, LIU Xiaohui¹^,⁴^,⁵, CHEN Danqiu¹^,⁴^,⁵, ZHU Jiaqi¹^,⁴^,⁵, WU Junqi¹^,⁴^,⁵

1. School of Earth and Space Sciences, Peking University, Beijing 100871, China
2. School of Electronics Engineering and Computer Science, Peking University, Beijing 100871, China
3. School of Software & Microelectronics, Peking University, Beijing 102600, China
4. National Experimental Teaching Demonstrating Center of Earth Sciences(Peking University), Beijing 100871, China
5. National Virtual Simulation Experimental Teaching Center of Earth Sciences(Peking University), Beijing 100871, China

Received:2020-03-30 Revised:2020-05-08 Online:2020-09-25 Published:2020-09-25

摘要/Abstract

摘要：

矿物识别在许多研究领域都有着重要作用,基于深度学习技术的智能矿物识别为这些领域带来了新的发展方向,不仅能有效节省人工成本,还能减小识别错误。针对石英、角闪石、黑云母、石榴石和橄榄石共5种矿物进行实验,提出了一种准确高效的智能矿物识别方法。实验采用图像分析常用的卷积神经网络建立模型,设计出一套基于残差神经网络的矿物识别方法。本实验独立采集了5种矿物的偏光显微图像数据集,用于模型的训练、验证和测试,并通过合理的数据增强策略来扩充训练数据集。在卷积神经网络的结构设计上,选取了ResNet-18作为框架,最终于模型测试中取得89%的准确率,成功训练出一个较为精准的矿物识别模型,实现了基于深度学习的智能矿物识别方法。

关键词: 深度学习, 矿物识别, 计算机视觉, 卷积神经网络, 残差神经网络

Abstract:

Mineral classification plays an important role in many research fields. Intelligent mineral identification based on deep learning brings a new development direction to these fields, it can effectively save labor costs as well as reducing classification errors. The purpose of this paper is to study an accurate, efficient and versatile intelligent mineral identification method by deep learning. We trained and tested this method on five kinds of minerals: quartz, hornblende, biotite, garnet and olivine. We used the convolution neural network, commonly applies to image analysis, to establish the model and designed the model structure based on residual network (ResNet). In order to support deep learning, we collected microscopic imaging data sets of five kinds of minerals independently, and used them to train, verify and test the model. Besides, we also expanded the data sets for training through reasonable data augmentation. In terms of structural design of the convolutional neural network, we selected ResNets-18 as the framework and finally trained a successful mineral identification model achieving 89% accuracy in the test.

Key words: deep learning, mineral classification, computer vision, convolutional neural network, residual neural network

中图分类号:

P57
TP391.41

郭艳军, 周哲, 林贺洵, 刘小辉, 陈丹丘, 祝佳琪, 伍峻琦. 基于深度学习的智能矿物识别方法研究[J]. 地学前缘, 2020, 27(5): 39-47.

GUO Yanjun, ZHOU Zhe, LIN Hexun, LIU Xiaohui, CHEN Danqiu, ZHU Jiaqi, WU Junqi. The mineral intelligence identification method based on deep learning algorithms[J]. Earth Science Frontiers, 2020, 27(5): 39-47.

图/表 14

表1 ResNet-18网络结构[25]

Table 1 Network structure of ResNet-18[25]

层名	输出大小	参数
conv1	112×112	7×7, 64, stride 2
conv2_x	56×56	3×3 max pool, stride 2
conv2_x	56×56	$3 × 3 64 3 × 3 64$ ×2
conv3_x	28×28	$3 × 3 128 3 × 3 128$ ×2
conv4_x	14×14	$3 × 3 256 3 × 3 256$ ×2
conv5_x	7×7	$3 × 3 512 3 × 3 512$ ×2
	1×1	average pool, 1 000-d fc, softmax
FLOPs		1.8×10⁹

表1 ResNet-18网络结构[25]

Table 1 Network structure of ResNet-18[25]

层名	输出大小	参数
conv1	112×112	7×7, 64, stride 2
conv2_x	56×56	3×3 max pool, stride 2
conv2_x	56×56	$3 × 3 64 3 × 3 64$ ×2
conv3_x	28×28	$3 × 3 128 3 × 3 128$ ×2
conv4_x	14×14	$3 × 3 256 3 × 3 256$ ×2
conv5_x	7×7	$3 × 3 512 3 × 3 512$ ×2
	1×1	average pool, 1 000-d fc, softmax
FLOPs		1.8×10⁹

图1 ResNet-18结构图

Fig.1 A structure chart of ResNet-18

图2 残差学习基本单元[25]

Fig.2 Basic unit of residual learning. Adapted from [25].

图3 数据处理流程图

Fig.3 Data processing flow chart

图4 模型训练及测试流程图

Fig.4 Model training and testing flow chart

图5 模型分类示例

Fig.5 Model classification example

图6 采集样本示例

Fig.6 Sample collection example

图7 基于图像翻转的数据增强

Fig.7 Data argumentation based on image flipping

图8 基于尺度变换的数据增强

Fig.8 Data argumentation based on scale transformation

图9 训练过程中 loss 变化

Fig.9 Loss changes during training

图10 验证集上准确度变化

Fig.10 Change of accuracy on validation set

表2 测试集得分

Table 2 Test set score

参数	矿物的识别情况
参数	角闪石	橄榄石	黑云母	石英	石榴石	总计
数量	20	20	20	20	20	100
正确数	20	15	16	18	20	89
准确率	100%	75%	80%	90%	100%	89%

表3 结果中易混淆的分析

Table 3 Confusing analysis of test results

实测矿物	预测概率
实测矿物	角闪石	橄榄石	黑云母	石英	石榴石
角闪石	1.00	0.00	0.00	0.00	0.00
橄榄石	0.00	0.75	0.00	0.15	0.10
黑云母	0.10	0.10	0.80	0.00	0.00
石英	0.00	0.10	0.00	0.90	0.00
石榴石	0.00	0.00	0.00	0.00	1.00

图11 错误案例展示以第1张图片为例,a表示案例序号,橄榄石表示样本的真实类别,括号中石榴石为模型识别出的类别,41%代表模型认为其是石榴石的概率。

Fig.11 Presentation of erred cases

参考文献 29

[1]	LU Y, YANG K, XIU L C, et al. Identification of hydrocarbon and clay minerals based on near-infrared spectroscopy and its geological significance[J]. Geological Bulletin of China, 2017,36(10):1884-1891.
[2]	WANG R S, YANG S M, BAI K. A Review of mineral spectral identification methods and models with imaging spectrometer[J]. Remote Sensing for Land Resources, 2007,19(1):1-9.
[3]	LIU S W, GAN F P, YAN B K, et al. Application of the imaging spectroscopic technique in mineral identification and mapping[J]. Geology in China, 2006,33(1):178-186.
[4]	MENG J K, HAO G, ZHOU Z L, et al. Application of multi-features matching decision tree to mineral identification[J]. Scientific and Technological Management of Land and Resources, 2012,29(6):93-96.
[5]	HINTON G E, OSINDERO S, TEH Y W. A fast learning algorithm for deep belief nets[J]. Neural Computation, 2006,18(7):1527-1554.
[6]	LECUN Y, BENGIO Y, HINTON G. Deep learning[J]. Nature, 2015,521(7553):436-444.
[7]	ZHOU Y Z, WANG J, ZUO R G, et al. Machine learning, deep learning and Python language in field of geology[J]. Acta Petrologica Sinica, 2018,34(11):3173-3178.
[8]	GIRSHICK R. Fast R-CNN[C]// Proceedings of 2015 IEEE international conference on computer vision (ICCV). Santiago: IEEE Computer Society, 2015: 1440-1448.
[9]	HE K, GKIOXARI G, DOLLAR P, et al. “Mask R-CNN”[C]// Proceedings of 2017 IEEE international conference on computer vision (ICCV). Venice: IEEE Computer Society, 2017: 2980-2988.
[10]	刘延保, 曹树刚, 刘玉成. 基于LS-SVM的岩石细观图像分析方法探讨[J]. 岩石力学与工程学报, 2008(5):1059-1065.
[11]	刘珏先, 滕奇志, 王正勇, 等. 基于协同表示的多特征融合岩石分类[J]. 计算机应用, 2016,36(3):854-858.
[12]	MARMO R, AMODIO S, TAGLIAFERRI R, et al. Textural identification of carbonate rocks by image processing and neural network: methodology proposal and examples[J]. Computers & Geosciences, 2005,31(5):649-659.
[13]	SINGH N, SINGH T N, TIWARY A, et al. Textural identification of basaltic rock mass using image processing and neural network[J]. Computational Geosciences, 2010,14(2):301-310.
[14]	徐述腾, 周永章. 基于深度学习的镜下矿石矿物的智能识别实验研究[J]. 岩石学报, 2018,34(11):3244-3252.
[15]	白林, 姚钰, 李双涛, 等. 基于深度学习特征提取的岩石图像矿物成分分析[J]. 中国矿业, 2018,27(7):178-182.
[16]	张野, 李明超, 韩帅. 基于岩石图像深度学习的岩性自动识别与分类方法[J]. 岩石学报, 2018,34(2):333-342.
[17]	ZHOU P, HAN X T, VLAD I, et al. Learning rich features for image manipulation detection[C]// Proceedings of 2018 IEEE conference on computer vision and pattern recognition (CVPR). Salt Lake City: IEEE Computer Society, 2018: 1053-1061.
[18]	Convolutional Neural Network. UFLDL Tutorial[EB/OL]. (2013-09-23)[2019-12-30]. http://ufldl.stanford.edu/tutorial/supervised/Exercise ConvolutionalNeuralNetwork/.
[19]	RUMELHART D E, HINTON G E, WILLIAMS R J, et al. Learning representations by back propagating errors[J]. Nature, 1986,323(6088):533-536.
[20]	LECUN Y, BOTTOU L. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998,86(11):2278-2324.
[21]	KRIZHEVSKY A, SUTSKEVER I, HINTON G E. Imagenet classification with deep convolutional neural networks[J]. Advances in Neural Information Processing Systems, 2012,25(2):1097-1105.
[22]	HINTON G E, SRIVASTAVA N, KRIZHEVSKY A, et al. Improving neural networks by preventing co-adaptation of feature detectors[J]. Computer Science, 2012,3(4):212-223.
[23]	SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[J]. ICLR, 2015.
[24]	SZEGEDY C, LIU W, JIA Y, et al. Going deeper with convolutions[C]// Proceedings of 2015 IEEE conference on computer vision and pattern recognition (CVPR). Boston: IEEE Computer Society, 2015: 1-9.
[25]	HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]// Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 770-778.
[26]	MICHIE D, SPIEGELHALTER D J, TAYLOR C C, et al. Machine learning, neural and statistical classification[M]. New York: Ellis Horwood/Upper Saddle River, NJ, USA: Prentice-Hall, 1995.
[27]	SEBASTINANI F. Machine learning in automated text categorization[J]. ACM Computing Surveys (CSUR), 2002,34(1):1-47.
[28]	SZEGEDY C, VANHOUCKE V, IOFFE S, et al. Rethinking the inception architecture for computer vision[C]// Proceedings of the IEEE conference on computer vision and pattern recognition. Seattle WA: IEEE Comp Soc; Comp Vis Fdn, 2016: 2818-2826.
[29]	SZEGEDY C, IOFFE S, VANHOUCKE V, et al. Inception-v4, inception-resnet and the impact of residual connections on learning[C]// Proceedings of the thirty-first AAAI conference on artificial intelligence. San Francisco, CA: Association of Advancement Artificial Intelligence, 2016: 1-12.

基于深度学习的智能矿物识别方法研究

The mineral intelligence identification method based on deep learning algorithms

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 14

参考文献 29

相关文章 1

编辑推荐

Metrics

本文评价