Workpiece Surface Defect Detection Based on Prototype Network With Blur Pooling

Owing to the development of computer vision, the research of workpiece surface defect detection obtain remarkable progress. However, the existing models are not well generalized in few-shot task, and are prone to over fitting. In this paper, A prototype network with blur pooling is proposed to satisfy Nyquist sampling theorem by adding a blur pooling to the feature extractor of the prototype network, which can suppress the interference of high-frequency noise in the original pictures and increase the generalization ability of the model. The experimental results on the Northeastern University (NEU) surface defect dataset and DAGM2007 dataset indicate that our approach outperforms conventional methods both on stability and accuracy.