Title:
DEFECT POINT IDENTIFICATION METHOD BASED ON DEEP LEARNING NETWORK
Document Type and Number:
WIPO Patent Application WO/2023/019636
Kind Code:
A1
Abstract:
A defect point identification method based on a deep learning network. The method first process an image having a defect by using a Resnet-50 feature extraction network, extracts features of the defect, and then detects and identifies the defect by using an enhanced transformer network. By enhancing the transformer network module of a DETR network, the method can detect the defect more accurately while improving the speed.
Inventors:
WANG HUIYAN (CN)
JIANG HUAN (CN)
JIANG HUAN (CN)
Application Number:
PCT/CN2021/115494
Publication Date:
February 23, 2023
Filing Date:
August 31, 2021
Export Citation:
Assignee:
UNIV ZHEJIANG GONGSHANG (CN)
International Classes:
G06N3/04; G06N3/08
Foreign References:
CN111899224A | 2020-11-06 | |||
CN111681228A | 2020-09-18 | |||
CN111260614A | 2020-06-09 | |||
US20190163904A1 | 2019-05-30 |
Attorney, Agent or Firm:
HANGZHOU AOCHUANG INTELLECTUAL PROPERTY AGENCY CO., LTD. (CN)
Download PDF: