Title:
METHOD FOR DETECTING DEFECT OF ITEM TO BE INSPECTED, DEVICE THEREFOR, AND COMPUTER PROGRAM THEREFOR
Document Type and Number:
WIPO Patent Application WO/2021/010430
Kind Code:
A1
Abstract:
Proposed is a highly accurate defect detection method using AI technology. This method for detecting a defect of an item to be inspected includes: a step for dividing, into a plurality of regions, a raw image that has captured an inspection subject together with a background; a valid region identification step for determining whether the inspection subject appears in a given divided region, and designating a region in which the inspection subject appears to be a valid region; a block-to-be-processed identification step for identifying a block to be processed which comprises a plurality of regions selected from the raw image and which includes at least one valid region; a defect identification step for detecting the presence/absence of a defect by implementing AI processing on the identified block to be processed, and if a defect is present, identifying the coordinates of the defect in the block to be processed and the probability thereof; a coordinates conversion step for converting the defect coordinates obtained by the defect identification step into raw coordinates of the raw image; and an inspection result creation step for creating a defect inspection result on the basis of the defect coordinates for the raw coordinates and the probability thereof. The block-to-be-processed identification step includes: a main block-to-be-processed identification step for identifying a block to be processed consisting of m x n (where m, n are natural numbers greater than 1) regions in the raw image; a first complementary block formation step for forming a first complementary block by causing the block to be processed to shift, by region units, in the direction in which m are arranged in the raw image; and a second complementary block formation step for forming a second complementary block by causing the block to be processed to shift, by region unit, in the direction in which n are arranged in the raw image. The defect identification step includes: a main defect identification step for identifying the coordinates of a defect in the block to be processed, and the probability thereof; a first complementary defect identification step for identifying the coordinates of a defect in the first complementary block, and the probability thereof; and a second complementary defect identification step for identifying the coordinates of a defect in the second complementary block, and the probability thereof. The coordinates conversion step converts, into raw coordinates of the raw image, the coordinates of the defects identified respectively in the main defect identification step, the first complementary defect identification step and the second complementary identification step.
More Like This:
Inventors:
ISHIGURO TAKAYUKI (JP)
Application Number:
PCT/JP2020/027584
Publication Date:
January 21, 2021
Filing Date:
July 16, 2020
Export Citation:
Assignee:
TKMES CORP (JP)
International Classes:
G01N21/88
Domestic Patent References:
| WO2018235266A1 | 2018-12-27 |
Foreign References:
| JP2010139317A | 2010-06-24 | |||
| JP2019028504A | 2019-02-21 | |||
| JP2019114927A | 2019-07-11 | |||
| JPH08152309A | 1996-06-11 | |||
| JP2005339180A | 2005-12-08 | |||
| JP2017166929A | 2017-09-21 | |||
| JP4657869B2 | 2011-03-23 |
Attorney, Agent or Firm:
KONISHI, Tomimasa (JP)
Download PDF: