US 12,423,804 B2
Image analysis apparatus and method for determining shape of particle included in image of object
Yoshishige Okuno, Tokyo (JP); Eriko Takeda, Tokyo (JP); Junichi Shibuya, Tokyo (JP); Yuki Nakao, Tokyo (JP); and Kenji Ogawa, Tokyo (JP)
Assigned to RESONAC CORPORATION, Tokyo (JP)
Filed by Resonac Corporation, Tokyo (JP)
Filed on Apr. 24, 2024, as Appl. No. 18/644,586.
Application 18/644,586 is a division of application No. 17/595,608, granted, now 11,995,816, previously published as PCT/JP2020/017799, filed on Apr. 24, 2020.
Claims priority of application No. 2019-098384 (JP), filed on May 27, 2019.
Prior Publication US 2024/0273711 A1, Aug. 15, 2024
Int. Cl. G06T 7/00 (2017.01); G01N 15/14 (2024.01); G01N 15/1434 (2024.01); G06T 7/60 (2017.01); G06V 10/74 (2022.01); G06V 10/75 (2022.01)
CPC G06T 7/001 (2013.01) [G01N 15/1434 (2013.01); G06T 7/60 (2013.01); G06V 10/752 (2022.01); G06V 10/761 (2022.01); G01N 2015/1497 (2013.01); G06T 2207/10056 (2013.01); G06T 2207/20084 (2013.01); G06T 2207/30136 (2013.01)] 12 Claims
OG exemplary drawing
 
1. An image analysis apparatus comprising:
a processor; and
a memory storing program instructions that cause the processor to:
determine a shape of a particle included in a particle image that is extracted from an image of an object, so that an OK particle image which is a particle image of an OK particle that satisfies a predetermined standard for shape and a provisional NG particle image which is a particle image of a provisional NG particle that does not satisfy the predetermined standard, are obtained;
prepare a generative model;
generate a pseudo image (fake) using the generative model; and
determine whether the provisional NG particle image and the pseudo image (fake) are similar, wherein in a case where the provisional NG particle image and the pseudo image (fake) are determined to be similar, the provisional NG particle is determined as including an OK particle, wherein
in the preparing of the generative model, machine learning is performed by using a network having an early-stage network in which the image of the object is input in an input layer and a feature is output in an output layer, and a late-stage network in which the feature output from the early-stage network is input in an input layer and a pseudo image generated by superimposing a plurality of particle images including the OK particle image is output in an output layer.