Patent · US Active

Method for defect classification, method for training defect classifier, and apparatus thereof

US11334982B2 · kind B2 · utility

1Cited by

2References

14Claims

0Family size

Assignees

Inventors

Xiangjun Peng · Beijing, CN
Yunqi Wang · Beijing, CN
Chenxi Zhao · Beijing, CN
Yachong Xue · Beijing, CN
Gang Li · Beijing, CN
Yaoyu Lv · Beijing, CN
Shuo Zhang · Oettingen, DE
Minglei Chu · Beijing, CN
Lili Chen · Beijing, CN
Hao Zhang · Shanghai, CN

Key dates

Filing date	Apr 27, 2020
Grant date	May 17, 2022
Priority date	—
Expiry date	May 9, 2040

Classification

Technology area (CPC G)Physics
CPC primaryG06T2207/30121
WIPO fieldComputer technology
WIPO sectorElectrical engineering

Abstract

The disclosure provides a method for defect classification, including: extracting a low-level feature of a defect region from a defect image; encoding the low-level feature by using a defect dictionary to obtain a middle-level semantic feature corresponding to the low-level feature; classifying a defect in the defect region of the defect image into one of a plurality of defect categories based on the middle-level semantic feature, wherein the defect dictionary includes a defect intra-category dictionary and a defect inter-category dictionary.

Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.