CADN: A weakly supervised learning-based category-aware object detection network for surface defect detection

作者：

Highlights：

• A novel Category-aware Conv-Pooling module is proposed, which explores weak image tag annotation to extract spatial information.

• Knowledge distillation strategy is adopted to force the feature of a student CADN to mimic that of a teacher CADN, leading to accuracy improvement.

• As verified, weakly supervised defect detection is achieved and competitive results are obtained by using the proposed CADN method.

• In CADN, human labeling effort, accuracy and speed are simultaneously considered, making the method practical in industrial applications.

摘要

•A novel Category-aware Conv-Pooling module is proposed, which explores weak image tag annotation to extract spatial information.•Knowledge distillation strategy is adopted to force the feature of a student CADN to mimic that of a teacher CADN, leading to accuracy improvement.•As verified, weakly supervised defect detection is achieved and competitive results are obtained by using the proposed CADN method.•In CADN, human labeling effort, accuracy and speed are simultaneously considered, making the method practical in industrial applications.

论文关键词：Weakly supervised learning,Automated surface inspection,Defect detection,Knowledge distillation

论文评审过程：Received 26 January 2020, Revised 17 June 2020, Accepted 31 July 2020, Available online 1 August 2020, Version of Record 11 August 2020.

论文官网地址：https://doi.org/10.1016/j.patcog.2020.107571