TY - JOUR
T1 - Defect detection in wrap film product using compact convolutional neural network
AU - Nakashima, Kento
AU - Nagata, Fusaomi
AU - Otsuka, Akimasa
AU - Watanabe, Keigo
AU - Habib, Maki K.
N1 - Publisher Copyright:
© 2021, International Society of Artificial Life and Robotics (ISAROB).
PY - 2021/8
Y1 - 2021/8
N2 - Although the automation of some inspection processes for various kinds of industrial products has progressed, the situation seems to be largely depending on visual inspection ability of inspectors who are familiar with the quality control of each product. Recently, not a few attempts have been tried to apply convolutional neural networks (CNNs) specialized in deep learning technology to image recognition for product defect detection. However, that of wrap film products is not easy due to, e.g., the reflection of light. In this paper, the authors introduce a CNN design tool to detect defects that appear in the manufacturing process of wrap film products. First, a template matching method is applied to the entire images of the wrap film products to extract only the target film part. Then, a compact CNN model is designed using the tool and trained using a large number of augmented images of good products and defective ones. Finally, the generalization ability of the CNN model is evaluated through classification experiments of test images, so that the desired accuracy over 0.95 could be achieved.
AB - Although the automation of some inspection processes for various kinds of industrial products has progressed, the situation seems to be largely depending on visual inspection ability of inspectors who are familiar with the quality control of each product. Recently, not a few attempts have been tried to apply convolutional neural networks (CNNs) specialized in deep learning technology to image recognition for product defect detection. However, that of wrap film products is not easy due to, e.g., the reflection of light. In this paper, the authors introduce a CNN design tool to detect defects that appear in the manufacturing process of wrap film products. First, a template matching method is applied to the entire images of the wrap film products to extract only the target film part. Then, a compact CNN model is designed using the tool and trained using a large number of augmented images of good products and defective ones. Finally, the generalization ability of the CNN model is evaluated through classification experiments of test images, so that the desired accuracy over 0.95 could be achieved.
KW - CNN design tool
KW - Convolutional neural network (CNN)
KW - Defect detection
KW - Visualization
KW - Wrap film products
UR - http://www.scopus.com/inward/record.url?scp=85107499899&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85107499899&partnerID=8YFLogxK
U2 - 10.1007/s10015-021-00686-y
DO - 10.1007/s10015-021-00686-y
M3 - Article
AN - SCOPUS:85107499899
VL - 26
SP - 360
EP - 366
JO - Artificial Life and Robotics
JF - Artificial Life and Robotics
SN - 1433-5298
IS - 3
ER -