Paw-Net: Stacking ensemble deep learning for segmenting scanning electron microscopy images of fine-grained shale samples
Abstract
Segmentation of scanning electron microscopy (SEM) images is critical yet time-consuming for geological analyses, as it needs to differentiate the boundaries for different mineral objects to facilitate subsequent analyses, such as porosity calculation. Recently, various machine learning methods, especially convolutional neural networks (CNNs), have been explored to segment SEM images of fine-grained shale samples. However, we found that general CNNs do not yield optimal performance due to insufficient training data and imbalanced objects in SEM images. This work has revised the U-Net architecture, a popular approach for biomedical image analyses, by incorporating a loss function that addresses the imbalance issue. Furthermore, we used the ensemble learning method to train multiple models and combined the results to improve the overall performance of segmentation. We prepared 2162 sub-images from raw SEM images in our experiments and divided them into training, validation, and testing datasets. The overall results show that our method improves the average Intersection over Union (IOU) of mineral objects from 0.49 to 0.58, compared to the original U-Net model. Our method can clearly distinguish each object from others with boundaries, even in highly imbalanced images. Training our models takes less than three minutes using a single GPU, while manual labelingmore »
- Authors:
-
- Univ. of Texas, Arlington, TX (United States)
- Rice Univ., Houston, TX (United States)
- Publication Date:
- Research Org.:
- Univ. of Texas, Arlington, TX (United States)
- Sponsoring Org.:
- USDOE Office of Nuclear Energy (NE)
- OSTI Identifier:
- 1894325
- Alternate Identifier(s):
- OSTI ID: 1884327
- Grant/Contract Number:
- NE0008797
- Resource Type:
- Accepted Manuscript
- Journal Name:
- Computers and Geosciences
- Additional Journal Information:
- Journal Volume: 168; Journal ID: ISSN 0098-3004
- Publisher:
- Elsevier
- Country of Publication:
- United States
- Language:
- English
- Subject:
- 58 GEOSCIENCES; Image segmentation; Neural network; Shale; Loss function; Ensemble learning; Imbalanced dataset
Citation Formats
Yin, Binqian, Hu, Qinhong, Zhu, Yingying, Zhao, Chen, and Zhou, Keren. Paw-Net: Stacking ensemble deep learning for segmenting scanning electron microscopy images of fine-grained shale samples. United States: N. p., 2022.
Web. doi:10.1016/j.cageo.2022.105218.
Yin, Binqian, Hu, Qinhong, Zhu, Yingying, Zhao, Chen, & Zhou, Keren. Paw-Net: Stacking ensemble deep learning for segmenting scanning electron microscopy images of fine-grained shale samples. United States. https://doi.org/10.1016/j.cageo.2022.105218
Yin, Binqian, Hu, Qinhong, Zhu, Yingying, Zhao, Chen, and Zhou, Keren. Tue .
"Paw-Net: Stacking ensemble deep learning for segmenting scanning electron microscopy images of fine-grained shale samples". United States. https://doi.org/10.1016/j.cageo.2022.105218. https://www.osti.gov/servlets/purl/1894325.
@article{osti_1894325,
title = {Paw-Net: Stacking ensemble deep learning for segmenting scanning electron microscopy images of fine-grained shale samples},
author = {Yin, Binqian and Hu, Qinhong and Zhu, Yingying and Zhao, Chen and Zhou, Keren},
abstractNote = {Segmentation of scanning electron microscopy (SEM) images is critical yet time-consuming for geological analyses, as it needs to differentiate the boundaries for different mineral objects to facilitate subsequent analyses, such as porosity calculation. Recently, various machine learning methods, especially convolutional neural networks (CNNs), have been explored to segment SEM images of fine-grained shale samples. However, we found that general CNNs do not yield optimal performance due to insufficient training data and imbalanced objects in SEM images. This work has revised the U-Net architecture, a popular approach for biomedical image analyses, by incorporating a loss function that addresses the imbalance issue. Furthermore, we used the ensemble learning method to train multiple models and combined the results to improve the overall performance of segmentation. We prepared 2162 sub-images from raw SEM images in our experiments and divided them into training, validation, and testing datasets. The overall results show that our method improves the average Intersection over Union (IOU) of mineral objects from 0.49 to 0.58, compared to the original U-Net model. Our method can clearly distinguish each object from others with boundaries, even in highly imbalanced images. Training our models takes less than three minutes using a single GPU, while manual labeling can take up to three hours for each image. Furthermore, the method helps geoscientists gain insights quickly and effectively by building neural network models from a small dataset of SEM images.},
doi = {10.1016/j.cageo.2022.105218},
journal = {Computers and Geosciences},
number = ,
volume = 168,
place = {United States},
year = {Tue Aug 23 00:00:00 EDT 2022},
month = {Tue Aug 23 00:00:00 EDT 2022}
}
Works referenced in this record:
Development of a 2D and 3D computational algorithm for discontinuity structural geometry identification by artificial intelligence based on image processing techniques
journal, June 2018
- Azarafza, Mohammad; Ghazifard, Akbar; Akgün, Haluk
- Bulletin of Engineering Geology and the Environment, Vol. 78, Issue 5
SegNet: A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation
journal, December 2017
- Badrinarayanan, Vijay; Kendall, Alex; Cipolla, Roberto
- IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 39, Issue 12
Breast Cancer Detection Using Extreme Learning Machine Based on Feature Fusion With CNN Deep Features
journal, January 2019
- Wang, Zhiqiong; Li, Mo; Wang, Huaxia
- IEEE Access, Vol. 7
Segmentation of digital rock images using deep convolutional autoencoder networks
journal, May 2019
- Karimpouli, Sadegh; Tahmasebi, Pejman
- Computers & Geosciences, Vol. 126
Secondary electron emission in the scanning electron microscope
journal, November 1983
- Seiler, H.
- Journal of Applied Physics, Vol. 54, Issue 11
Multimodal Biometric Authentication Systems Using Convolution Neural Network Based on Different Level Fusion of ECG and Fingerprint
journal, January 2019
- Hammad, Mohamed; Liu, Yashu; Wang, Kuanquan
- IEEE Access, Vol. 7
A comprehensive review on soil classification using deep learning and computer vision techniques
journal, January 2021
- Srivastava, Pallavi; Shukla, Aasheesh; Bansal, Atul
- Multimedia Tools and Applications, Vol. 80, Issue 10
The generalized sigmoid activation function: Competitive supervised learning
journal, June 1997
- Narayan, Sridhar
- Information Sciences, Vol. 99, Issue 1-2
A survey of thresholding techniques
journal, February 1988
- Sahoo, P. K.; Soltani, S.; Wong, A. K. C.
- Computer Vision, Graphics, and Image Processing, Vol. 41, Issue 2
Atomic number and crystallographic contrast images with the SEM: a review of backscattered electron techniques
journal, March 1987
- Lloyd, Geoffrey E.
- Mineralogical Magazine, Vol. 51, Issue 359
Traffic sign detection and recognition using fully convolutional network guided proposals
journal, November 2016
- Zhu, Yingying; Zhang, Chengquan; Zhou, Duoyou
- Neurocomputing, Vol. 214
Deep learning-based method for SEM image segmentation in mineral characterization, an example from Duvernay Shale samples in Western Canada Sedimentary Basin
journal, May 2020
- Chen, Zhuoheng; Liu, Xiaojun; Yang, Jijin
- Computers & Geosciences, Vol. 138
Morphology, Genesis, and Distribution of Nanometer-Scale Pores in Siliceous Mudstones of the Mississippian Barnett Shale
journal, November 2009
- Loucks, R. G.; Reed, R. M.; Ruppel, S. C.
- Journal of Sedimentary Research, Vol. 79, Issue 12
Intelligent Identification for Rock-Mineral Microscopic Images Using Ensemble Machine Learning Algorithms
journal, September 2019
- Zhang, Ye; Li, Mingchao; Han, Shuai
- Sensors, Vol. 19, Issue 18
Land suitability evaluation using image processing based on determination of soil texture–structure and soil features
journal, January 2020
- Mahmoodi‐Eshkaftaki, Mahmood; Haghighi, Amin; Houshyar, Ehsan
- Soil Use and Management, Vol. 36, Issue 3
Deep learning-based landslide susceptibility mapping
journal, December 2021
- Azarafza, Mohammad; Azarafza, Mehdi; Akgün, Haluk
- Scientific Reports, Vol. 11, Issue 1
Nano-scale pore structure and fractal dimension of organic-rich Wufeng-Longmaxi shale from Jiaoshiba area, Sichuan Basin: Investigations using FE-SEM, gas adsorption and helium pycnometry
journal, February 2016
- Yang, Rui; He, Sheng; Yi, Jizheng
- Marine and Petroleum Geology, Vol. 70
Spacing and block volume estimation in discontinuous rock masses using image processing technique: a case study
journal, July 2021
- Azarafza, Mohammad; Koçkar, Mustafa K.; Faramarzi, Lohrasb
- Environmental Earth Sciences, Vol. 80, Issue 14
Low nanopore connectivity limits gas production in Barnett formation
journal, December 2015
- Hu, Qinhong; Ewing, Robert P.; Rowe, Harold D.
- Journal of Geophysical Research: Solid Earth, Vol. 120, Issue 12
Properties of cross-entropy minimization
journal, July 1981
- Shore, J.; Johnson, R.
- IEEE Transactions on Information Theory, Vol. 27, Issue 4
Remote Sensing Image Scene Classification Using CNN-CapsNet
journal, February 2019
- Zhang, Wei; Tang, Ping; Zhao, Lijun
- Remote Sensing, Vol. 11, Issue 5