Semi-supervised anomaly detection for visual quality inspection

作者：

Highlights：

• A pre-trained neural network is blended with a statistical-based transformation.

• The blended network is able to remove any anomalies from the input image.

• The method requires few samples for training even though is based on deep learning.

• Very fast domain adaptation based only on anomaly-free samples.

摘要

•A pre-trained neural network is blended with a statistical-based transformation.•The blended network is able to remove any anomalies from the input image.•The method requires few samples for training even though is based on deep learning.•Very fast domain adaptation based only on anomaly-free samples.

论文关键词：Quality control,Visual quality inspection,Anomaly detection,Computer vision,Convolutional neural networks,CNNs

论文评审过程：Received 18 December 2020, Revised 22 April 2021, Accepted 22 May 2021, Available online 8 June 2021, Version of Record 11 June 2021.

论文官网地址：https://doi.org/10.1016/j.eswa.2021.115275