矿物组分识别与智能解释在不同岩性之间的信息共享与迁移学习

doi:10.13745/j.esf.sf.2024.5.8

摘要/Abstract

摘要：

在地球科学领域,岩石微观观测数据的采集过程繁琐且效率低下,这不仅增加了研究成本,降低了可靠性,同时也限制了数据的开源共享。此外,由于岩性的多样性和观测手段的差异,单一数据集的规模通常较小,这对于依赖大规模数据集的深度学习框架而言是一大挑战。为此,本研究探索迁移学习如何促进不同岩性间的信息共享,并通过此机制提高矿物组分识别与智能解释任务的模型性能。通过采集不同区域、岩性、矿物组分和偏光模式下的铸体薄片样本,本文深入研究了深度学习模型在不同观测对象和手段下的迁移学习机制,并聚焦于探索地质信息的深层表征。研究成果不但揭示了迁移学习在促进地质学领域信息共享与模型性能提升中的关键作用,还为自动化和智能化地质认识融合奠定了基础。实验结果显示,通过迁移学习,本文模型在智能解释任务中的准确率显著提高,从53.3%提高至98.73%,而在矿物组分识别任务中,准确率也实现了近10%的提升。这些成果证明了迁移学习在地质学领域内解决实际问题和提高模型泛化能力、性能和稳定性方面的巨大潜力。

关键词: 迁移学习, 薄片矿物组分识别, 薄片图像智能解释, 地质认识融合

Abstract:

In earth sciences the rock microscopic data collection process is both labor-intensive and inefficient, which have a negative impact on research cost/reliability and open data sharing. Additionally, rock heterogeneity and variation in data collection methods typically result in small-scale datasets—this poses a significant challenge to deep learning frameworks that rely on large-scale datasets for training. To address this issue, we investigate how transfer learning can facilitate information sharing across different rock types and enhance model performance in tasks such as mineral identification and intelligent interpretation. By compiling thin section image datasets, taking in diverse rock sampling regions, rock types, mineral compositions, viewed under varying viewing modes, we delve into the mechanisms of transfer learning across different observational targets and methods, focusing on the deep representation of geological information. Our findings not only highlight the pivotal role of transfer learning in promoting information sharing and improving model performance within the field of geosciences, but also lay a foundation for the automatic and intelligent integration of geological insights. According to experimental results, transfer learning led to significant accuracy improvement, from 53.3% to 98.73%, in intelligent interpretation task, and a nearly 10% improvement in mineral identification task. These results convincingly showcase the great potential of transfer learning in addressing practical problems in geology as well as enhancing model generalization, model performance, and model stability.

Key words: transfer learning, thin section mineral composition identification, thin section image intelligent interpretation, geological understanding integration

中图分类号:

刘烨, 韩雨伯, 朱文瑞. 矿物组分识别与智能解释在不同岩性之间的信息共享与迁移学习[J]. 地学前缘, 2024, 31(4): 95-111.

LIU Ye, HAN Yubo, ZHU Wenrui. Mineral component identification and intelligent interpretation: Information sharing and transfer learning across different lithologies[J]. Earth Science Frontiers, 2024, 31(4): 95-111.

图/表 30

图1 原始数据集 a—A数据集,砂岩,红色铸体薄片,单偏光; b—B数据集,片麻岩,常规薄片,单偏光;c—B数据集,片麻岩,常规薄片,正交偏光。

Fig.1 Sample of thin section images in datasets A (a) and B (b, c)

图2 总体方法流程图

Fig.2 Research methodology flowchart

图3 数据集预处理的目的与意义

Fig.3 Dataset preprocessing scheme

图4 单体岩石数据集预处理流程

Fig.4 Preprocessing workflow for monolith rock dataset

图5 部分数据集A-a示例 a—高岭石;b—孔隙;c—岩屑;d—石英;e—杂基;f—其他。

Fig.5 Sample of thin section images in dataset A-a

图6 部分数据集B-a示例 a—钾长石;b—斜长石;c—石英。

Fig.6 Sample of thin section images in dataset B-a

表1 组分识别数据集组成及相关参数

Table 1 Description of component recognition datasets A and B

数据集	数据集A(图像)						数据集B(图像)
(预处理)组分分类	A-a						B-a
类别	石英	高岭石	岩屑	杂基	孔隙	其他	石英	斜长石	钾长石
数量	1 200	1 200	1 200	1 200	1 200	1 200	50	50	50

图7 部分数据集C示例

Fig.7 Sample of thin section images in dataset C

图8 部分数据集D示例

Fig.8 Sample of thin section images in dataset D

图9 迁移学习方法流程

Fig.9 Flowchart of transfer learning method

图10 预训练组分识别模型

Fig.10 Pre-trained component recognition model

图11 基于预训练模型的迁移模型

Fig.11 Transfer learning model based on pre-trained model

图12 卷积块注意力模块

Fig.12 Convolutional block attention module

图13 智能解释语义模型

Fig.13 Semantic model used in intelligent interpretation

图14 语义模型编码器

Fig.14 Semantic model encoder

图15 语义模型解码器

Fig.15 Semantic model decoder

图16 GRU训练过程

Fig.16 GRU training process

表2 组分识别模型超参数

Table 2 Hyperparameters in component recognition model

模型	预训练模型						组分识别迁移模型
超参数	学习率	批大小	迭代次数	优化器	损失函数	学习率衰减	学习率	批大小	迭代次数	优化器	损失函数	学习率衰减
	0.003	16	80	Adam	交叉熵损失	每30次迭代将学习率乘以0.5	0.000 3	8	40	Adam	交叉熵损失	每20次迭代将学习率乘以0.5

表3 智能解释语义模型超参数

Table 3 Hyperparamerters in semantic model

模型	智能解释语义预训练模型
超参数	字典长度	最大语句长度	嵌入层尺寸	GRU隐藏层尺寸	学习率	批大小	迭代次数	优化器	损失函数	学习率衰减
	30	32	16	512	0.000 2	64	150	Adam	交叉熵损失	每5次迭代将学习率乘以0.1
模型	智能解释语义迁移模型
超参数	字典长度	最大语句长度	嵌入层尺寸	GRU隐藏层尺寸	学习率	批大小	迭代次数	优化器	损失函数	学习率衰减
	16	18	16	512	0.0002	16	50	Adam	交叉熵损失	每5次迭代将学习率乘以0.1

图17 预训练模型训练准确率和测试准确率变化

Fig.17 Evolution of the predictive accuracy of pre-trained component recognition model during training (blue) and testing (green)

图18 数据集A-a预训练模型混淆矩阵 a—训练集混淆矩阵;b—测试集混淆矩阵。

Fig.18 Confusion matrices for pre-trained component recognition model on dataset A-a

图19 组分识别迁移模型和原始模型训练对照 a—训练集损失函数对照;b—测试集损失函数对照。

Fig.19 Comparisons of component recognition model performances between transfer learning model (blue) and the original model (orange) during training (a) and testing (b)

表4 模型预测准确率对比

Table 4 Comparison of the predictive accuracy of semantic model with (blue) and without (orange) transfer learning

模型	类别	准确率/%
原始模型	训练集	86.6
原始模型	测试集	87.5
迁移模型	训练集	92.5
迁移模型	测试集	96.8

图20 数据集B-a通过迁移模型和原始模型的混淆矩阵 a—迁移模型混淆矩阵;b—原始模型混淆矩阵。

Fig.20 Confusion matrices for component recognition model on dataset B-a with (a) and without (b) transfer learning

图21 数据集A样本组分识别结果 a—数据集A中样本;b—样本待识别区域;c—组分识别结果。

Fig.21 Component recognition results on dataset A

图22 数据集B样本组分识别结果 a—数据集B中正交偏光样本;b—数据集B中单偏光样本;c—正交偏光样本待识别区域;d—单偏光样本待识别区域;e—组分识别结果。

Fig.22 Component recognition results on dataset B

图23 智能解释预训练模型生成解释结果

Fig.23 Semantic interpretation results using pre-trained model

图24 智能解释迁移模型和原始模型训练对照 a—训练集损失函数对照;b—测试集损失函数对照。

Fig.24 Comparisons of semantic model performances between transfer learning model (blue) and the original model (orange) during training (a) and testing (b)

表5 模型生成描述准确率对比

Table 5 Comparison of the descriptive accuracy of semantic model with and without transfer learning

模型	类别	准确率/%
原始模型	训练集	57.6
原始模型	测试集	53.3
迁移模型	训练集	100
迁移模型	测试集	98.3

图25 智能解释迁移模型生成解释结果

Fig.25 Semantic interpretation results using transfer learning model

参考文献 23

[1]	周永章, 张良均, 张奥多, 等. 地球科学大数据挖掘与机器学习[M]. 广州: 中山大学出版社, 2018.
[2]	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.
[3]	程国建, 郭文惠, 范鹏召. 基于卷积神经网络的岩石图像分类[J]. 西安石油大学学报(自然科学版), 2017, 32(4): 116-122.
[4]	白林, 魏昕, 刘禹, 等. 基于VGG模型的岩石薄片图像识别[J]. 地质通报, 2019, 38(12): 2053-2058.
[5]	谭永健, 田苗, 徐德馨, 等. 基于Xception网络的岩石图像分类识别研究[J]. 地理与地理信息科学, 2022, 38(3): 17-22.
[6]	POLAT Ö, POLAT A, EKICI T. Automatic classification of volcanic rocks from thin section images using transfer learning networks[J]. Neural Computing and Applications, 2021, 33(18): 11531-11540.
[7]	姜枫. 基于语义识别的砂岩薄片图像分割方法研究[D]. 南京: 南京大学, 2018.
[8]	QIAO W D, ZHAO Y F, XU Y, et al. Deep learning-based pixel-level rock fragment recognition during tunnel excavation using instance segmentation model[J]. Tunnelling and Underground Space Technology, 2021, 115: 104072.
[9]	YIN B Q, HU Q H, ZHU Y Y, et al. Paw-Net: stacking ensemble deep learning for segmenting scanning electron microscopy images of fine-grained shale samples[J]. Computers and Geosciences, 2022, 168: 105218.
[10]	刘烨, 吕锦涛. 基于超像素与半监督的岩石图像分割与识别[J]. 工程科学与技术, 2023, 55(2): 171-183.
[11]	廖启俊. 基于递归网络的图文标注算法研究[D]. 广州: 华南理工大学, 2017.
[12]	SHI Y L, YANG W Z, DU H X, et al. Overview of image captions based on deep learning[J]. Acta Electonica Sinica, 2021, 49(10): 2048-2060.
[13]	PAN S J, YANG Q. A survey on transfer learning[J]. IEEE Transactions on Knowledge and Data Engineering, 2010, 22(10), 1345-1359.
[14]	SHIN H C, ROTH H R, GAO M, et al. Deep convolutional neural networks for computer-aided detection: CNN architectures, dataset characteristics and transfer learning[J]. IEEE Transactions On Medical Imaging, 2016, 35(5): 1285-1298.
[15]	TAN C, SUN F, KONG T, et al. A survey on deep transfer learning: with an emphasis on domain adaptation techniques[C]// Artificial neural networks and machine learning-ICANN 2018: 27th international conference on Artificial Neural Networks. Rhodes: Springer International Publishing, 2018: 270-279.
[16]	TSCHANNEN V, DELESCLUSE M, RODRIGUEZ M, et al. Facies classification from well logs using an inception convolutional network[EB/OL]. (2017-06-02)[2024-01-15]. https://doi.org/10.48550/arXiv.1706.00613.
[17]	PIRES DE LIMA R, SURIAMIN F, MARFURT K J, et al. Convolutional neural networks as aid in core lithofacies classification[J]. Interpretation, 2019, 7(3): SF27-SF40.
[18]	KOESHIDAYATULLAH A, MORSILLI M, LEHRMANN D J, et al. Fully automated carbonate petrography using deep convolutional neural networks[J]. Marine and Petroleum Geology, 2020, 122: 104687.
[19]	DAWSON H L, DUBRULE O, JOHN C M. Impact of dataset size and convolutional neural network architecture on transfer learning for carbonate rock classification[J]. Computers and Geosciences, 2023, 171: 105284.
[20]	ACHANTA R, SHAJI A, SMITH K, et al. SLIC superpixels compared to state-of-the-art superpixel methods[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2012, 34(11): 2274-2282. PMID
[21]	SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[EB/OL].(2014-09-04)[2024-01-16]. https://doi.org/10.48550/arXiv.1409.1556.
[22]	WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module[M]//Computer Vision - ECCV 2018. Cham: Springer International Publishing, 2018: 3-19.
[23]	CHUNG J, GULCEHRE C, CHO K H, et al. Empirical evaluation of gated recurrent neural networks on sequence modeling[EB/OL]. (2014-12-11)[2024-01-21]. https://doi.org/10.48550/arXiv.1412.3555.