US 12,094,180 B2
Method for developing machine-learning based tool
Chandra Suwandi Wijaya, Singapore (SG); and Ariel Beck, Singapore (SG)
Assigned to PANASONIC INTELLECTUAL PROPERTY MANAGEMENT CO., LTD., Osaka (JP)
Filed by PANASONIC INTELLECTUAL PROPERTY MANAGEMENT CO., LTD., Osaka (JP)
Filed on Oct. 6, 2020, as Appl. No. 17/063,923.
Prior Publication US 2022/0108210 A1, Apr. 7, 2022
Int. Cl. G06N 20/00 (2019.01); G06F 11/34 (2006.01); G06F 18/214 (2023.01); G06F 18/231 (2023.01); G06V 10/25 (2022.01); G06V 10/40 (2022.01)
CPC G06V 10/25 (2022.01) [G06F 11/3409 (2013.01); G06F 18/214 (2023.01); G06F 18/231 (2023.01); G06N 20/00 (2019.01); G06V 10/40 (2022.01)] 16 Claims
OG exemplary drawing
 
10. A system for developing machine-learning (ML) based tool for visual inspection, the system comprising:
an initialization module for initializing a visual inspection process based on one or more selected images defining a dataset;
an alignment model aligning one or more detected objects within an image-frame of at least one selected image of the one or more selected images of at least selected one image of the one or more selected images to generate at least one aligned image of the at least one selected image and to train the ML based tool using the generated at least one aligned image;
a feature extractor for extracting a plurality of features from the at least one aligned image;
an ML training module for receiving a selection of a machine-learning classifier to operate upon the extracted features and classify the at least one selected image using one or more labels pre-defined under a visual inspection; and
a metacontroller for communicating with one or more of the alignment model, the feature extractor and the ML training module for assessing one or more of:
a performance of at least one of first model and the feature extractor;
a comparison of operation among the one or more selected machine-learning classifiers; and
diagnosis of an unexpected operation with respect to one or more of the first model, the feature extractor and the selected machine-learning classifier,
wherein the metacontroller performs an automatic or semi-automatic A/B testing based on observing a change in configuration of a pipeline having stages as one or more of the first model, second model, feature extractor and the ML training module.