Verfahren zur Erzeugung eines Basismodells für Maschinenlernmodelle zur Fehlererkennung und Verfahren zur Erzeugung eines Maschinenlernmodells zur Fehlererkennung

Details

Veröffentlichungs-Nr.

EP4814888

Anmeldetag

13. März 2026

Veröffentlichung

30. September 2026

Rechtsraum

EP

IPC

G06V10/762, G06V10/778

Abstract

A method of creating a basic model for multiple machine learning models corresponding to multiple defect detection criteria. The method includes the steps of preparing first learning data (Da) (S101), extracting feature values from respective image data items (Di) of the first learning data (S102), performing clustering of the image data items in accordance with the feature values (S103), extracting some image data items-in accordance with the clustering result (S 104 to S 107), and creating a basic model (Mb) by performing learning using basic data (Db) (S108). The basic data includes data on common defectives identified as defectives in common on all of the defect detection criteria. By performing individual learning processing corresponding to the individual defect detection criteria on the basic model (Mb), it is possible to efficiently create a model with desired accuracy. That is, machine learning models corresponding to the individual defect detection criteria can be created with high accuracy and efficiency.

Anmelder

Firma

SCREEN Holdings Co., Ltd.

Land

🇯🇵 Japan

Fachgebiete