Multi-scale channel enhanced transformer for rock thin sections identification and sequence consistency optimization.
Saved in:
| Title: | Multi-scale channel enhanced transformer for rock thin sections identification and sequence consistency optimization. |
|---|---|
| Authors: | Guo, Xiaoyao1,2 (AUTHOR) xiaoyao1206@foxmail.com, Chen, Yan1,2 (AUTHOR) carly_chen@126.com, He, Shipeng3 (AUTHOR) heshipeng@petrochina.com.cn, Zhang, Xingpeng1,2 (AUTHOR) xpzhang@swpu.edu.cn, Zhou, Jing1 (AUTHOR) shenwangxiaobai@163.com, Bao, Xucheng1,2 (AUTHOR) xucheng.bao@foxmail.com |
| Source: | Computational Geosciences. Jun2025, Vol. 29 Issue 3, p1-19. 19p. |
| Subjects: | Convolutional neural networks, Image recognition (Computer vision), Transformer models, Deep learning, Brewster's angle |
| Abstract: | The identification of rock thin sections plays a pivotal role in geological exploration, as it provides critical insights into the fundamental properties and composition of rocks. However, the accurate identification of mineral particles presents significant challenges due to three primary factors: the inherent imbalance in data distribution, misclassification caused by feature similarity, and substantial feature variations observed under different cross-polarized angles. These complexities render conventional deep-learning models with single-structure architectures inadequate for precise mineral particle identification. Therefore, this study proposes a novel rock thin section image classification methodology that combines a Multi-Scale Channel Enhanced Transformer (MSCET) with a Sequence Consistency Optimization (SCO) strategy. This integrated approach is designed to effectively extract distinctive features of mineral particles while fully exploiting the influence of polarization angle variations. The MSCET architecture synergistically combines Convolutional Neural Networks (CNN), Squeeze-and-Excitation Networks (SENet), and Transformer mechanisms to enhance the network's feature representation capabilities. Specifically, it employs distinct convolutional operations to extract both coarse- and fine-grained features of mineral particles. The SENet and Transformer structures are then utilized to aggregate global information across both channel and spatial dimensions. Furthermore, we introduce the SCO strategy to refine low-confidence predictions, thereby mitigating the impact of feature variations in multi-angle cross-polarized images. Comprehensive experimental evaluations demonstrate the efficacy of our proposed method, achieving a classification accuracy of 92.35% on the test set. The method also shows significant improvements in key performance metrics, including recall, precision, and F1 score, substantiating its potential for robust rock thin section identification in geological applications. [ABSTRACT FROM AUTHOR] |
| Copyright of Computational Geosciences is the property of Springer Nature and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.) | |
| Database: | Engineering Source |
Be the first to leave a comment!
