Related Experiment Video
Updated: Jan 11, 2026
06:37
Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
Published on: December 15, 2023
5.2K
Measured multi-source semi-supervised working condition recognition based on curvelet pooling and attention mechanism
Shuo Yang1, Bin Zhou2, Yanjiang Wang3
1School of Computer Science and Technology, Shandong University of Technology, Zibo, 255000, People's Republic of China.
Scientific Reports
|November 14, 2025
Summary
This study introduces an advanced deep learning method for recognizing oil well working conditions using curvelet pooling and multi-source data fusion. The technique improves accuracy and practicality by leveraging unlabeled data for better performance with limited labeled samples.
Area of Science:
- Petroleum Engineering
- Artificial Intelligence
- Image Processing
Background:
- Sucker-rod pumping wells generate massive image data from multiple sources.
- Accurate identification of working conditions is crucial for efficient oil extraction.
- Existing methods may struggle with the complexity and volume of multi-source data.
Purpose of the Study:
- To develop a novel, accurate, and practical method for oil well working condition recognition.
- To enhance the capability of deep learning models in processing complex, multi-source image data.
- To improve classification performance and generalization using semi-supervised learning with abundant unlabeled data.
Main Methods:
- Curvelet pooling optimization integrated into ResNet-50 for enhanced feature extraction.
- Multi-source attention mechanism fusion for combining ground dynamometer and electrical power card data.
- Semi-supervised deep learning classification with dynamic pseudo-label confidence and class fairness regularization.
Main Results:
- The proposed method efficiently processes multi-source data from sucker-rod pumping wells.
- It demonstrates improved performance compared to traditional deep learning frameworks.
- The technique effectively utilizes unlabeled data to enhance recognition accuracy and practical applicability.
Conclusions:
- The developed method offers a significant advancement in oil well working condition recognition.
- It successfully integrates curvelet transform, attention mechanisms, and semi-supervised learning.
- The approach enhances engineering practicability by minimizing the need for labeled training samples.
