Related Experiment Video
Updated: Jun 20, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
A Zero-Shot Learning Approach for Blockage Detection and Identification Based on the Stacking Ensemble Model.
Chaoqun Li1, Zao Feng1,2, Mingkai Jiang3
1Faculty of Information and Automation, Kunming University of Science and Technology, Kunming 650500, China.
Sensors (Basel, Switzerland)
|September 14, 2024
Summary
This study introduces a novel zero-shot pipeline blockage detection method using stacking ensemble learning. The approach identifies new defects without prior training data, achieving 72.5% accuracy for unknown defect categories.
Area of Science:
- Engineering
- Machine Learning
- Signal Processing
Background:
- Data-driven defect identification relies heavily on labeled samples.
- Emerging defects during data acquisition create a scarcity of labeled data for new defect types.
Purpose of the Study:
- To develop a zero-shot learning method for pipeline blockage detection and identification.
- To address the challenge of identifying novel defects without specific training examples.
Main Methods:
- Variational Modal Decomposition (VMD) for signal decomposition.
- Information entropy calculation on intrinsic modal functions (IMFs) to create feature sets.
- Stacking ensemble attribute learning using an attribute matrix for defect categories.
Main Results:
- Successful identification of target defects without prior training samples.
- Enhanced classification performance across six experimental datasets.
- An average recognition accuracy of 72.5% for previously unknown defect categories.
Conclusions:
- The proposed zero-shot learning method effectively identifies pipeline blockages, even for novel defects.
- Stacking ensemble learning combined with VMD and information entropy offers a robust solution for defect identification with limited data.
