公开(公告)号：US20230162344A1

公开(公告)日：2023-05-25

申请号：US17958474

申请日：2022-10-03

Applicant: Keyence Corporation

Inventor： Xinliang ZHAO

IPC: G06T7/00 ,  G06T5/00 ,  G06V10/75

CPC classification number: G06T7/0008 ,  G06T5/002 ,  G06V10/751 ,  G06T2207/10024 ,  G06T2207/20081 ,  G06T2207/20182

Abstract: When a machine learning network trained with both a non-defective product image and a defective product image is used, it is possible to stably exhibit high detection capability for the defective product image having an unknown defect while shortening a takt time during the operation time. A processor executes a first learning process of causing a machine learning network to learn a non-defective product image added with a noise, and a second learning process of causing the machine learning network to learn a defective product image, and detects both an unknown defect having a characteristic different from a characteristic of the non-defective product image and a known defect having a characteristic designated as a defective site, by inputting a workpiece image to the machine learning network whose parameter has been adjusted by the first learning process and the second learning process.