Image recognition device and method

Patent metadata

Field	Value
Publication number	US-12014530-B2
Application number	US-201817286604-A
Country	US
Kind code	B2
Filing date	Dec 21, 2018
Priority date	Dec 21, 2018
Publication date	Jun 18, 2024
Grant date	Jun 18, 2024

Abstract

In order to select an optimal learning model for an image when inference is carried out in the extraction of a profile line using machine learning, without requiring a correct value or degree of certainty, a feature extraction learning model group containing a plurality of learning models is used for feature extraction. A recall learning model group containing recall learning models is paired with the feature extraction learning models. A feature amount extraction unit for referencing a feature extraction learning model and extracting a feature amount from input data; a data-to-data recall unit for referencing a recall learning model and outputting a recall result with the feature amount subjected to dimensional compression; and a learning model selection unit for selecting a feature extraction learning model from the feature extraction learning model group under the condition that the difference between the feature amount and the recall result is minimized are provided.

First claim

The invention claimed is:

1. An image recognition device, comprising:
a feature extraction learning model group that stores a plurality of feature extraction learning models;
a recall learning model group that stores a recall learning model which is paired with a feature extraction learning model among the plurality of feature extraction learning models;
a feature amount extraction unit that extracts a feature amount from input data with reference to the feature extraction learning model;
a data-to-data recall unit that outputs a recall result accompanied by dimensional compression of the feature amount with reference to the recall learning model; and
a learning model selection unit that selects the feature extraction learning model from a feature extraction learning model group under a condition that a difference between the feature amount and the recall result is minimized.

2. The image recognition device according to claim 1 , comprising:
a learning model suitability determination unit that determines whether the feature extraction learning model is suitable, the feature extraction learning model being selected for a population where a sample of the input data is sampled, from a difference between the feature amount and the recall result.

3. The image recognition device according to claim 2 ,
wherein, when the learning model suitability determination unit determines that the feature extraction learning model is not suitable, the feature extraction learning model is reselected using a sample of the input data.

4. The image recognition device according to claim 1 , comprising:
a training data creation support unit that includes a supervised user interface for narrowing down an input place when a difference between the feature amount and the recall result is large in the sample of the input data; and
a learning model learning unit that learns the feature extraction learning model using training data created by the training data creation support unit.

5. The image recognition device according to claim 4 ,
wherein there is provided a function of roughly drawing the input data in the user interface in the training data creation support unit to draw a category of the feature amount, and to input the category of the feature amount.

6. The image recognition device according to claim 4 ,
wherein the training data creation support unit performs at least one of obtaining the input place using the plurality of feature amounts and recall results and switching the input place.

7. The image recognition device according to claim 4 ,
wherein the learning model learning unit further learns the recall learning model, and
wherein the feature amount learning model learned by the learning model learning unit is added to the feature extraction learning model group, and the recall learning model learned by the learning model learning unit is added to the feature extraction learning model group.

8. The image recognition device according to claim 1 ,
wherein the feature amount is a category of an element in the input data.

9. The image recognition device according to claim 1 ,
wherein the input data is an image, and the feature amount is a contour line or a design drawing.

10. The image recognition device according to claim 1 ,
wherein the dimensional compression is performed by using principal component analysis or an autoencoder.

11. The image recognition device according to claim 1 ,
wherein one or more feature amount extraction units using a method other than machine learning are included in the feature amount extraction unit.

12. The image recognition device according to claim 1 ,
wherein the learning model selection unit displays one or more of a selection result of the feature extraction learning model, the difference, and a selection range of the feature extraction learning model in a screen.

13. An image recognition device, comprising:
a feature extraction learning model group that stores a plurality of feature extraction learning models;
a feature amount extraction unit that extracts a feature amount from input data with reference to a feature extraction learning model among the plurality of feature extraction learning models; and
a learning model selection unit that calculates a common scale capable of being compared among a plurality of types of learning models from a score when the feature amount extraction unit extracts the feature amount, and selects the feature extraction learning model using the common scale from a feature extraction learning model group.

14. The image recognition device according to claim 13 , comprising:
a learning model suitability determination unit that determines whether the feature extraction learning model selected from the common scale is suitable.

15. The image recognition device according to claim 14 , comprising:
a learning model reselection unit that reselects the feature extraction learning model using a sample of the input data when the learning model suitability determination unit determines that it is not suitable.

16. The image recognition device according to claim 13 , comprising:
a training data creation support unit that includes a supervised user interface for narrowing down an input place when the common scale is small in a sample of the input data; and
a learning model learning unit that learns the feature extraction learning model using training data created by the training data creation support unit.

17. The image recognition device according to claim 16 ,
wherein the user interface in the training data creation support unit has a function of roughly drawing the input data to draw a category of the feature amount, and to input a category of the feature amount.

18. The image recognition device according to claim 16 ,
wherein the feature amount learning model learned by the learning model learning unit is added to a feature extraction learning model group.

19. The image recognition device according to claim 13 ,
wherein the feature amount is a category of an element in the input data.

20. The image recognition device according to claim 13 ,
wherein the input data is an image, and the feature amount is a contour line or a design drawing.

21. The image recognition device according to claim 13 ,
wherein the common scale is a statistic representing a degree of variation in the score or a statistic representing a degree of protrusion of the score.

22. The image recognition device according to claim 13 ,
wherein the common scale is a correct answer rate converted from the score.

23. The image recognition device according to claim 13 ,
wherein one or more feature amount extraction units using a method other than machine learning are included in the feature amount extraction unit.

24. The image recognition device according to claim 13 ,
wherein the learning model selection unit displays one or more of a selection result of the feature extraction learning model, a difference, and a selection range of the feature extraction learning model in a screen.

25. An image recognition method that has a plurality of feature extraction learning models and a plurality of recall learning models which are paired with a feature extraction learning model among the plurality of feature extraction learning models, comprising:
extracting a feature amount from input data with reference to the feature extraction learning model;
obtaining a recall r

Assignees

• Hitachi High Tech Corp

Inventors

• Yumiba Ryou
• Toyoda Yasutaka
• Shindo Hiroyuki

Classifications

• G06F18/213

  Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods · CPC title

• G06V20/698

  Matching; Classification · CPC title

• G06V20/695

  Preprocessing, e.g. image segmentation · CPC title

• G06V10/82

  using neural networks · CPC title

• G06V10/7715

  Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods · CPC title