Related Experiment Video
Updated: Sep 13, 2025
11:54
Microfluidic Platform with Multiplexed Electronic Detection for Spatial Tracking of Particles
Published on: March 13, 2017
9.4K
Crossed Wavelet Convolution Network for Few-Shot Defect Detection of Industrial Chips
Zonghai Sun1, Yiyu Lin1, Yan Li1
1School of Automation Science and Engineering, South China University of Technology, Guangzhou 510641, China.
Sensors (Basel, Switzerland)
|July 30, 2025
Summary
This study introduces a novel crossed wavelet convolution network (CWCN) for detecting defects in resistive polymer humidity sensor chips. The CWCN method effectively improves few-shot learning (FSL) performance in mixed-scene industrial conditions.
Area of Science:
- Materials Science
- Electrical Engineering
- Computer Science
Background:
- Resistive polymer humidity sensor performance relies heavily on resistor chip quality.
- Detecting chip defects is difficult due to limited defective samples, hindering traditional supervised learning.
- Existing few-shot learning (FSL) methods face challenges with mixed-scene conditions in industrial defect detection.
Purpose of the Study:
- To develop an effective few-shot learning method for industrial chip defect detection.
- To address the limitations of current FSL approaches in handling mixed-scene data.
- To improve the accuracy and robustness of defect detection in sensor chip manufacturing.
Main Methods:
- Proposed a crossed wavelet convolution network (CWCN) incorporating a dual-pipeline crossed wavelet convolution training framework (DPCWC).
- Introduced ProSL, a novel loss value calculation module utilizing prototype learning.
- Applied wavelet transform convolution for feature fusion across multiple scenarios.
Main Results:
- CWCN demonstrated superior detection quality in few-shot tasks on chip datasets.
- Achieved mean average precision (mAP) improvements ranging from 2.76% to 16.43% over existing FSL methods.
- Validation on the open-source NEU-DET dataset confirmed the method's effectiveness.
Conclusions:
- The proposed CWCN method effectively fuses multi-scenario feature information for enhanced industrial defect detection.
- CWCN significantly improves few-shot learning performance in challenging mixed-scene environments.
- This approach offers a promising solution for quality control in sensor chip manufacturing.
