Battery Panel Defect Detection Method Based on Deep Convolutional Neural Network

Defect detection of product surface can be made manually or automatically utilizing pattern recognition. Traditional methods usually suffer from low efficiency, high cost, sensitivity to environmental changes, and low detection accuracy. This paper proposes a deep convolutional neural network (DCNN)...

Full description

Saved in:
Bibliographic Details
Published in2019 11th International Conference on Wireless Communications and Signal Processing (WCSP) pp. 1 - 6
Main Authors Jiang, Shibao, Wang, Taotao, Zhang, Shengli, Wang, Wei, Wang, Hui
Format Conference Proceeding
LanguageEnglish
Published IEEE 01.10.2019
Subjects
Classification algorithms
Convolutional neural networks
Data models
deep convolutional neural network
defect detection
Entropy
Feature extraction
focal loss
Machine learning
Training
unbalanced data classification
Online AccessGet full text

Cover

More Information
Summary:Defect detection of product surface can be made manually or automatically utilizing pattern recognition. Traditional methods usually suffer from low efficiency, high cost, sensitivity to environmental changes, and low detection accuracy. This paper proposes a deep convolutional neural network (DCNN) approach for battery panel defect detection. The training data can be collected by taking images for battery panels from an actual production line. Usually, this training data suffers from many problems such as poor image quality, underrepresentation of defective samples, and different types of defects. We use data enhancement techniques to preprocess the acquired images to obtain as many different samples as possible. Furthermore, we utilized the improved Focal Loss function to deal with prominent problems such as small samples and non-uniform sample data sets. The method achieved good results in the experiment.
ISSN:2472-7628
DOI:10.1109/WCSP.2019.8928008