File Download

There are no files associated with this item.

  Links for fulltext
     (May Require Subscription)
Supplementary

Article: Automatic, Point-Wise Rock Image Enhancement by Novel Unsupervised Deep Learning: Dataset Establishment and Model Development

TitleAutomatic, Point-Wise Rock Image Enhancement by Novel Unsupervised Deep Learning: Dataset Establishment and Model Development
Authors
KeywordsConvolutional neural network (CNN)
Deep curve estimation (DCE)
Rock image dataset establishment
Rock image light enhancement
Unsupervised deep learning
Issue Date26-Aug-2023
PublisherSpringer
Citation
Rock Mechanics and Rock Engineering, 2023, v. 56, n. 11, p. 8503-8541 How to Cite?
Abstract

Rock images play a vital role in providing data for engineering geological studies. However, low-light (a.k.a. dark) rock images are often obtained, and the existing methods for image enhancement, whether they are software-based, algorithm-based, or data-driven-based, are incapable of addressing the issue. This paper establishes a novel unsupervised deep learning (DL) model that caters to low-light rock image enhancement. The light enhancement process is carried out automatically and pixel-wise by utilizing the deep curve estimation (DCE) algorithm and a convolutional neural network (CNN). This study establishes a rock image dataset, encompassing diverse (1) rock types (tuff, siltstone, & granite groups), (2) light levels (2400 lx, 1200 lx, &100 lx), (3) color temperatures (5500 K, 4200 K, & 3000 K), and (4) surface conditions (wet & dry). The novel DL model is developed based on (1) high-order curves of the DCE algorithm, (2) an elaborate CNN architecture with step-wise convolutions and a squeeze and excitation module, (3) three non-reference loss functions, and (4) an alerting level for better generalization. The DL model possesses several advantageous features: (1) automatic light enhancement of rock images without subjective inputs, (2) independence from paired data, which requires manual retouching, (3) pixel-wise adjustment of intensities across color channels, (4) preservation of rock details without distortion by the attention mechanism, (5) high inference speeds, e.g., ~ 3 ms per rock image, and (6) most importantly, state-of-the-art performance when compared to 10 other widely-used enhancement methods, both visually and quantitatively, by three evaluation indices. Despite being designed and trained for rock images, the DL model also demonstrates impressive performance on non-rock images.


Persistent Identifierhttp://hdl.handle.net/10722/340498
ISSN
2023 Impact Factor: 5.5
2023 SCImago Journal Rankings: 1.902
ISI Accession Number ID

 

DC FieldValueLanguage
dc.contributor.authorZhou, Yimeng-
dc.contributor.authorWong, Louis Ngai Yuen-
dc.date.accessioned2024-03-11T10:45:05Z-
dc.date.available2024-03-11T10:45:05Z-
dc.date.issued2023-08-26-
dc.identifier.citationRock Mechanics and Rock Engineering, 2023, v. 56, n. 11, p. 8503-8541-
dc.identifier.issn0723-2632-
dc.identifier.urihttp://hdl.handle.net/10722/340498-
dc.description.abstract<p>Rock images play a vital role in providing data for engineering geological studies. However, low-light (a.k.a. dark) rock images are often obtained, and the existing methods for image enhancement, whether they are software-based, algorithm-based, or data-driven-based, are incapable of addressing the issue. This paper establishes a novel unsupervised deep learning (DL) model that caters to low-light rock image enhancement. The light enhancement process is carried out automatically and pixel-wise by utilizing the deep curve estimation (DCE) algorithm and a convolutional neural network (CNN). This study establishes a rock image dataset, encompassing diverse (1) rock types (tuff, siltstone, & granite groups), (2) light levels (2400 lx, 1200 lx, &100 lx), (3) color temperatures (5500 K, 4200 K, & 3000 K), and (4) surface conditions (wet & dry). The novel DL model is developed based on (1) high-order curves of the DCE algorithm, (2) an elaborate CNN architecture with step-wise convolutions and a squeeze and excitation module, (3) three non-reference loss functions, and (4) an alerting level for better generalization. The DL model possesses several advantageous features: (1) automatic light enhancement of rock images without subjective inputs, (2) independence from paired data, which requires manual retouching, (3) pixel-wise adjustment of intensities across color channels, (4) preservation of rock details without distortion by the attention mechanism, (5) high inference speeds, e.g., ~ 3 ms per rock image, and (6) most importantly, state-of-the-art performance when compared to 10 other widely-used enhancement methods, both visually and quantitatively, by three evaluation indices. Despite being designed and trained for rock images, the DL model also demonstrates impressive performance on non-rock images.<br></p>-
dc.languageeng-
dc.publisherSpringer-
dc.relation.ispartofRock Mechanics and Rock Engineering-
dc.subjectConvolutional neural network (CNN)-
dc.subjectDeep curve estimation (DCE)-
dc.subjectRock image dataset establishment-
dc.subjectRock image light enhancement-
dc.subjectUnsupervised deep learning-
dc.titleAutomatic, Point-Wise Rock Image Enhancement by Novel Unsupervised Deep Learning: Dataset Establishment and Model Development-
dc.typeArticle-
dc.identifier.doi10.1007/s00603-023-03490-1-
dc.identifier.scopuseid_2-s2.0-85169122781-
dc.identifier.volume56-
dc.identifier.issue11-
dc.identifier.spage8503-
dc.identifier.epage8541-
dc.identifier.eissn1434-453X-
dc.identifier.isiWOS:001063646200001-
dc.identifier.issnl0723-2632-

Export via OAI-PMH Interface in XML Formats


OR


Export to Other Non-XML Formats