System for creating learned model for component image recognition, and method for creating learned model for component image recognition

Patent metadata

Field	Value
Publication number	US-11386546-B2
Application number	US-201816963988-A
Country	US
Kind code	B2
Filing date	Feb 9, 2018
Priority date	Feb 9, 2018
Publication date	Jul 12, 2022
Grant date	Jul 12, 2022

Abstract

A system for creating a learned model for component image recognition, the learned model being used when performing image recognition of a component that is picked up by a suction nozzle of a component mounter or a component that is mounted on a circuit board, serving as an imaging target, by imaging the imaging target with a camera, and the system includes a computer configured to acquire a reference-learned model to be used for image recognition of a reference component. The computer collects sample component images for each type of a component having a predetermined similarity with the reference component, and creates a component-by-component learned model to be used for image recognition of the component for each type of the component by adding the sample component image for each type of the component, as teacher data of the reference-learned model, and re-learning the added sample component image.

First claim

The invention claimed is:

1. A system for creating a learned model for component image recognition, the learned model being used when performing image recognition by imaging an imaging target with a camera, the imaging target being a component that is picked up by a suction nozzle of a component mounter, or a component that is mounted on a circuit board, the system comprising:
a computer configured to acquire a reference-learned model to be used for image recognition of a reference component,
wherein the computer is configured to
collect sample component images for each type of a component having a predetermined similarity with the reference component,
obtain information of an inspection result acquired from an inspection device,
calculate a failure occurrence rate for each type of the component based on the information of the inspection result,
when the failure occurrence rate exceeds a predetermined threshold, create a component-by-component learned model to be used for image recognition of the component for each type of the component, the component-by-component learned model being created by adding the sample component image for each type of the component as teacher data of the reference-learned model, and re-learning the added sample component image.

2. The system for creating a learned model for component image recognition according to claim 1 , wherein the component-by-component learned model created for each type of the component by the computer is included in image processing component shape data prepared for each type of the component.

3. The system for creating a learned model for component image recognition according to claim 1 , wherein the component having a predetermined similarity with the reference component is a component having the same or similar shape even if any of a size, color, material, manufacturing company, and manufacturing lot of the component is different from the reference component.

4. The system for creating a learned model for component image recognition according to claim 1 , wherein the computer collects an image obtained by imaging the imaging target with a camera of a component mounter or a camera of the inspection device during production, as the sample component image.

5. The system for creating a learned model for component image recognition according to claim 1 , wherein the reference-learned model and the component-by-component learned model are learned models for determining whether a pickup orientation of the component picked up by the suction nozzle is normal pickup or abnormal pickup.

6. The system for creating a learned model for component image recognition according to claim 1 , wherein the reference-learned model and the component-by-component learned model are learned models for determining a presence or absence of the component picked up by the suction nozzle.

7. The system for creating a learned model for component image recognition according to claim 1 , wherein the reference-learned model and the component-by-component learned model are learned models for determining a presence or absence of the component mounted on the circuit board.

8. The system for creating a learned model for component image recognition according to claim 1 , wherein the computer transfers the created component-by-component learned model to a component mounter or the inspection device that uses the component-by-component learned model.

9. A method for creating a learned model for component image recognition, the learned model being used when performing image recognition by imaging the imaging target with a camera, the imaging target being a component that is picked up by a suction nozzle of a component mounter, or a component that is mounted on a circuit board, the method comprising:
acquiring a reference-learned model to be used for image recognition of a reference component;
collecting sample component images for each type of a component having a predetermined similarity with the reference component;
obtaining information of an inspection result acquired from an inspection device;
calculating a failure occurrence rate for each type of the component based on the information of the inspection result; and
when the failure occurrence rate exceeds a predetermined threshold, creating a component-by-component learned model to be used for image recognition of the component for each type of the component, the component-by-component learned model being created by adding the sample component image acquired for each type of the component as teacher data of the reference-learned model, and re-learning the added sample component image.

Assignees

• Fuji Corp

Inventors

• Kobayashi Takahiro
• Kito Shuichiro
• Yokoi yuta

Classifications

• G06F18/2155

  characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling · CPC title

• G06T2207/30148

  Semiconductor; IC; Wafer · CPC title

• G06T2207/20081

  Training; Learning · CPC title

• H05K13/08

  Monitoring manufacture of assemblages · CPC title

• G06T7/001

  using an image reference approach · CPC title