Learning Data Collection Device Image Suitability Determination

Solution Overview

Problem

The collection of training data for image recognition is inefficient due to the inclusion of improper images, which reduces recognition accuracy, and the process of labeling data is labor-intensive.

Innovation Solution

A learning data collection device and system that acquires images, determines their suitability as training data, and prompts the image capturer to reshoot if the image is not suitable, using various determination criteria and notification modes to ensure proper image capture.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If training data is collected by capturing images with various shooting conditions, then the quantity of training data increases, but the proportion of improper images increases reducing accuracy

Engineering Contradiction:

Improvequantity of training dataVSAvoidaccuracy of training data

Core Design Contradiction:

Quantity of substanceVSManufacturing precision

Solution Approach 1:

The system performs preliminary determination of image suitability immediately after image capture, before the image is fully processed or labeled. By evaluating whether the captured image meets training data requirements at this early stage, the system prevents improper images from entering the training dataset, thus maintaining data accuracy while enabling continuous data collection.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system implements a feedback mechanism where the determination result of image suitability is immediately fed back to control whether the captured image is registered as training data. This closed-loop control ensures that only images meeting the suitability criteria are added to the training dataset, automatically maintaining data quality without requiring manual review of each image.

Inventive Principle:

Principle #23Feedback

2Manufacturing precision

If all captured images are manually reviewed and labeled, then the accuracy of training data selection improves, but the time and labor required increases significantly

Engineering Contradiction:

Improveaccuracy of training data selectionVSAvoidtime for labeling data

Core Design Contradiction:

Manufacturing precisionVSLoss of time

Solution Approach 1:

The system performs automatic determination of image suitability using computational algorithms that evaluate whether captured images meet training data requirements. This self-service approach replaces manual review with automated assessment, significantly reducing the time and labor required while maintaining consistent evaluation criteria. The system serves itself by automatically filtering appropriate training data without human intervention for each image.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The patent replaces the mechanical process of manual image review and labeling with an automated determination system that uses computational methods to assess image suitability. This substitution eliminates the need for human operators to manually evaluate each captured image, dramatically reducing time consumption and labor requirements while maintaining or improving selection accuracy through consistent algorithmic evaluation.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Productivity

If improper images are included in training data, then data collection efficiency improves, but recognition accuracy deteriorates

Engineering Contradiction:

Improvedata collection efficiencyVSAvoidrecognition accuracy

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The system performs preliminary determination of image suitability immediately after image capture, before the image is fully processed or labeled. By evaluating whether the captured image meets training data requirements at this early stage, the system prevents improper images from entering the training dataset, thus maintaining data accuracy while enabling continuous data collection.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system implements a feedback mechanism where the determination result of image suitability is immediately fed back to control whether the captured image is registered as training data. This closed-loop control ensures that only images meeting the suitability criteria are added to the training dataset, automatically maintaining data quality without requiring manual review of each image.

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS11579904B2Learning data collection device, learning data collection system, and learning data collection method

Publication Date: 2023.02.14 PANASONIC INTELLECTUAL PROPERTY MANAGEMENT CO LTD

AI summary

In collection of training data for image recognition, in order to support a reduction in collection of improper images which are not suitable as training data, a learning data collection device includes a processor which is configured to acquire a captured image from an image capturing device, determine whether or not the captured image is suitable as training data, and when the captured image is determined to be not suitable as training data, perform a notification operation to prompt an image capturing person to reshoot a new image for the captured image.