Related Experiment Video
Updated: May 28, 2025
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
1.7K
PPLA-Transformer: An Efficient Transformer for Defect Detection with Linear Attention Based on Pyramid Pooling.
Xiaona Song1, Yubo Tian1, Haichao Liu1
1School of Mechanical Engineering, North China University of Water Resources and Electric Power, Zhengzhou 450045, China.
Sensors (Basel, Switzerland)
|February 13, 2025
Summary
This study introduces an efficient deep learning model for industrial defect detection, enhancing accuracy and speed. The new method improves upon Swin-Transformer, offering better performance with lower computational cost for quality control.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Industrial Quality Control
Background:
- Subtle defects in industrial products challenge traditional quality control, demanding high accuracy and efficiency.
- Deep learning models, like Swin-Transformer, show promise but face computational burdens limiting their industrial application.
- Existing methods struggle to balance detection accuracy with the high-speed requirements of industrial settings.
Purpose of the Study:
- To develop a novel deep learning model for industrial surface defect detection that enhances both accuracy and efficiency.
- To address the computational limitations of Swin-Transformer in industrial defect detection tasks.
- To improve the extraction of both global and local features for more precise defect identification.
Main Methods:
- Proposed a novel model incorporating a linear attention mechanism with pyramid pooling to reduce computational load and improve efficiency.
- Integrated partial convolution to enhance local feature extraction and further boost detection precision.
- Evaluated the model on a self-constructed SIM card slot defect dataset and the public PKU-Market-PCB dataset.
Main Results:
- The proposed model achieved superior performance compared to Swin-Transformer on both datasets.
- Demonstrated significant improvements in mean Average Precision (mAP) and Frames Per Second (FPS) with minimal computational cost.
- Outperformed Swin-Transformer by 1.2% mAP and 52 FPS on the SIM card slot dataset and 1.7% mAP and 51 FPS on the PKU-Market-PCB dataset.
Conclusions:
- The developed linear attention mechanism with pyramid pooling effectively enhances industrial defect detection accuracy and efficiency.
- The model's ability to improve performance while reducing computational cost validates its practical applicability in industrial quality control.
- The proposed approach shows universality and significant advantages over existing Swin-Transformer models for surface defect detection.
