基于多尺度区域卷积神经网络的露头孔洞自动提取

doi:10.19657/j.geoscience.1000-8527.2021.003

摘要/Abstract

摘要：

碳酸盐岩储层解释与评价的一个重要内容是确定其孔隙空间特征,野外露头是地下储层的真实刻画,对野外露头进行孔洞的自动化提取与其参数定量表征具有重要意义。在深度学习Mask-RCNN模型基础上进行改进,提出一种基于多尺度的区域卷积神经网络孔洞检测新方法,并通过两种方式进行准确度分析,第一种是将该方法的孔洞提取结果与OSTU分割法、分水岭分割法、BP神经网络法、支持向量机法以及Mask-RCNN的孔洞提取结果进行比较,其结果显示该方法有更高的检测准确度;第二种是通过该方法提取的孔洞结果计算洞数量、面孔率和洞面积均值三个孔洞特征参数,以人工提取结果为参照,比较得到洞数量准确度在88%以上,面孔率准确度在93%以上,洞面积均值准确度在93%以上。最后将提出的方法应用于峨边先锋灯二段的数字露头剖面孔洞自动识别,并分层计算孔洞参数,定量分析其分布特征,为该露头碳酸盐岩储层评价提供了依据。

关键词: 孔洞自动识别, 卷积神经网络, 数字露头, 深度学习

Abstract:

Determination of the pore space characteristics is important for carbonate reservoir interpretation and evaluation. Field outcrop can reflect the geology of the underground reservoir, and thus can be used to identify the cavity automatically and characterize their parameters. In this study, through enhancing the deep-learning model Mask-RCNN, a new cavity detection method based on a multi-scale regional convolution neural network is proposed, with its accuracy being verified by two methods: (1) By comparing the cavity extraction results of this method with those of OSTU, watershed, BP neural network, support vector machine, and Mask-RCNN, it is shown that the method has higher detection accuracy; (2) By calculating the three cavity characteristic parameters of cavity number, surface porosity, and the average cavity area through the cavity results extracted by the method, and by comparing the results of manual extraction, it is shown that the accuracy for the cavity number, surface porosity, and average cavity area is over 88%, 93%, and 93%, respectively. Consequently, the proposed method is applied to the automatic cavity identification in the digital outcrop profile of Dengying Formation (2^nd Member) in Xianfeng, Ebian. We calculated the cavity parameters in the layers, and quantitatively analyze their distribution characteristics, in order to provide a carbonate reservoir evaluation basis for this outcrop.

Key words: cavity automatic recognition, convolution neural network, digital outcrop, deep learning

中图分类号:

TE122.2

王庆, 曾齐红, 张友焱, 邵燕林, 魏薇, 邓帆. 基于多尺度区域卷积神经网络的露头孔洞自动提取[J]. 现代地质, 2021, 35(04): 1147-1154.

WANG Qing, ZENG Qihong, ZHANG Youyan, SHAO Yanlin, WEI Wei, DENG Fan. Automatic Extraction of Outcrop Cavity Based on Multi-scale Regional Convolution Neural Network[J]. Geoscience, 2021, 35(04): 1147-1154.

图/表 11

参考文献 25

[1]	刘学锋, 马乙云, 曾齐红, 等. 基于数字露头的地质信息提取与分析:以鄂尔多斯盆地上三叠统延长组杨家沟剖面为例[J]. 岩性油气藏, 2015, 27(5):13-18.
[2]	曾齐红, 马乙云, 谢兴, 等. 鄂尔多斯盆地延长组数字露头表层建模方法研究[J]. 岩性油气藏, 2015, 27(5):25-29.
[3]	李雪英, 蔺景龙, 文慧俭. 碳酸盐岩孔洞空间的自动识别[J]. 大庆石油学院学报, 2005, 9(4):4-6.
[4]	柯式镇, 许淑霞. 井壁电成像测井资料定量评价方法研究[J]. 天然气工业, 2006, 26(9):62-64.
[5]	曹飞. 裂缝性岩石声波参数实验研究及裂缝性储层测井评价[D]. 长春:吉林大学, 2015.
[6]	谢丹丹. 碳酸盐岩缝洞型储层参数定量评价方法研究[D]. 武汉:长江大学, 2015.
[7]	田金文, 高谦, 杜拥军, 等. 基于井壁成像测井图像的溶洞自动检测方法[J]. 江汉石油学院学报, 1999, 21(2):20-22.
[8]	李茂兵. 电成像测井自动识别和定量评价研究[D]. 青岛:中国石油大学(华东),2010.
[9]	申辉林, 高松洋. 基于BP神经网络进行裂缝识别研究[J]. 断块油气田, 2007, 14(2):60-62.
[10]	瞿子易, 周文, 罗鑫, 等. 基于粒子群和支持向量机的裂缝识别[J]. 石油与天然气地质, 2009, 30(6):786-792.
[11]	石广仁. 支持向量机在裂缝预测及含气性评价应用中的优越性[J]. 石油勘探与开发, 2008, 35(5):588-594.
[12]	陈钢花, 胡琮, 曾亚丽, 等. 基于BP 神经网络的碳酸盐岩储层缝洞充填物测井识别方法[J]. 石油物探, 2015, 54(1):99-104.
[13]	苏培东, 秦启荣, 黄润秋. 储层裂缝预测研究现状与展望[J]. 西南石油学院学报, 2005, 27(5):14-17.
[14]	HE K, ZHANG X, REN S, et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 37(9):1904-1916. DOI URL
[15]	GIRSHICK R. Fast RcCNN[M]//IEEE. 2015 IEEE International Conference on Computer Vision (ICCV).Santiago:IEEE, 2015: 1440-1448.
[16]	REN S, HE K, GIRSHICK R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017, 39(6):1137-1149.
[17]	DAI J F, LI Y, HE K M, et al. R-FCN: Object detection via region-based fully convolutional networks[EB/OL]. 2016: arxiv: 1605.06409[cs.CV]. https://arxiv.org/abs/1605.06409 .
[18]	HE K M, GKIOXARI G, DOLLAR P, et al. Mask R-CNN[M]//IEEE. IEEE International Conference on Computer Vision (ICCV). Venice:IEEE, 2017:2961-2969.
[19]	REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: Unified, real-time object detection[M]//IEEE. IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas, NV, USA: IEEE, 2016: 779-788.
[20]	LIU W, ANGUELOV D, ERHAN D, et al. SSD: Single Shot MultiBox Detector[M]. Cham: Springer International Publishing, 2016: 21-37.
[21]	HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[M]//IEEE. IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas, NV, USA: IEEE, 2016: 770-778.
[22]	FAN H, GUI S X, JINGWEN H, et al. Transferring deep convolutional neural networks for the scene classification of high-resolution remote sensing imagery[J]. Remote Sensing, 2015, 7(11):14680-14707. DOI URL
[23]	LIN T Y, MAIRE M, BELONGIE S, et al. Microsoft COCO: Common Objects in Context[M]. Cham: Springer International Publishing, 2014: 740-755.
[24]	DUTTA A, ZISSERMAN A. The VIA Annotation Software for Images, Audio and Video[M]. New York, USA: ACM, 2019: 2276-2279.
[25]	张泽苗, 霍欢, 赵逢禹. 深层卷积神经网络的目标检测算法综述[J]. 小型微型计算机系统, 2019, 40(9):1825-1831.

方法	精准率	召回率	准确率	F1-Score
OSTU	0.045 8	0.934 8	0.825 8	0.087 4
分水岭	0.067 0	0.513 7	0.931 9	0.118 5
BP神经网络	0.210 5	0.693 8	0.974 1	0.322 9
支持向量机	0.266 5	0.607 6	0.981 6	0.370 5
Mask-RCNN	0.587 2	0.537 1	0.989 2	0.561 0
我们的模型	0.700 1	0.749 1	0.994 9	0.723 8

方法	精准率	召回率	准确率	F1-Score
OSTU	0.045 8	0.934 8	0.825 8	0.087 4
分水岭	0.067 0	0.513 7	0.931 9	0.118 5
BP神经网络	0.210 5	0.693 8	0.974 1	0.322 9
支持向量机	0.266 5	0.607 6	0.981 6	0.370 5
Mask-RCNN	0.587 2	0.537 1	0.989 2	0.561 0
我们的模型	0.700 1	0.749 1	0.994 9	0.723 8

方法	精准率	召回率	准确率	F1-Score
OSTU	0.044 1	0.913 2	0.828 4	0.084 2
分水岭	0.055 9	0.434 2	0.931 8	0.099 1
BP神经网络	0.253 2	0.479 7	0.983 3	0.331 4
支持向量机	0.291 3	0.437 8	0.985 9	0.349 8
Mask-RCNN	0.426 9	0.387 2	0.986 8	0.406 1
我们的模型	0.607 5	0.631 7	0.993 3	0.619 4

方法	精准率	召回率	准确率	F1-Score
OSTU	0.044 1	0.913 2	0.828 4	0.084 2
分水岭	0.055 9	0.434 2	0.931 8	0.099 1
BP神经网络	0.253 2	0.479 7	0.983 3	0.331 4
支持向量机	0.291 3	0.437 8	0.985 9	0.349 8
Mask-RCNN	0.426 9	0.387 2	0.986 8	0.406 1
我们的模型	0.607 5	0.631 7	0.993 3	0.619 4

主要参数	实验区域一				实验区域二
主要参数	模型	人工	准确度/%		模型		人工		准确度/%
洞数量	28	25	88.00	43		39		89.74
面孔率/%	0.95	0.89	93.26	0.88		0.86		97.67
洞面积均值 /mm²	224.10	238.22	94.07	131.69		140.79		93.54