Inspection Image Selection Using Correction History for Retraining
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing region detectors trained by machine learning require extensive time and unsuitable learning data for efficient retraining, making it difficult to improve detection accuracy effectively.
Innovation Solution
A learning data collection apparatus and method that quantifies correction histories to extract inspection images with significant user corrections, excluding minor corrections, for efficient retraining of region detectors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If web crawlers are used to collect training data from websites, then large amounts of data can be collected automatically, but the data quality is low and contains many duplicates and errors
Solution Approach 1:
The patent introduces an intermediary system consisting of a server and client device that acts as a mediator between the web crawler and the training data collection process. The server receives images from the crawler, performs quality assessment, and manages the collection process, thereby improving data quality while maintaining automated collection efficiency
Solution Approach 2:
The patent implements feedback mechanisms where the system evaluates the quality of collected images and uses this information to adjust the collection process. The server assesses image quality metrics and provides feedback to control whether images are saved or rejected, ensuring continuous improvement of data quality
2Adaptability or versatility
If diverse data sources are collected to improve model adaptability, then the training data becomes more varied, but data quality control becomes more difficult
Solution Approach 1:
The patent applies local quality control by assessing and managing the quality of data from different sources individually. The server evaluates images based on their specific characteristics and source, applying appropriate quality standards to each data source while maintaining overall data quality consistency
Solution Approach 2:
The patent changes quality assessment parameters based on the data source and collection context. The system adjusts evaluation criteria dynamically according to the specific source characteristics, allowing effective quality control across diverse sources while maintaining adaptability
3Measurement precision
If manual data collection methods are used to ensure high data quality, then accurate and relevant data can be obtained, but the collection process is time-consuming and labor-intensive
Solution Approach 1:
The patent segments the data collection process into distinct automated and manual components. The automated crawler handles initial data gathering while the server performs automated quality filtering, reserving manual intervention only for edge cases, thereby reducing overall collection time while maintaining quality
Solution Approach 2:
The system implements self-service quality control where the server automatically assesses and filters images based on predefined quality criteria. This automated self-evaluation reduces the need for manual quality checking while maintaining high data quality standards
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
Provided are a learning data collection apparatus, a learning data collection method, and a program for collecting learning data to be used for efficient retraining. A learning data collection apparatus (10) includes an inspection image acquisition unit (11) that acquires an inspection image, a region detection result acquisition unit (damage detection result acquisition unit (13)) that acquires a region detection result the region detection result indicating a region detected by a region detector that is trained, a correction history acquisition unit (15) that acquires a correction history of the region detection result, a calculation unit (17) that calculates correction quantification information obtained by quantifying the correction history, a database that stores the inspection image, the region detection result, and the correction history in association with each other, an image extraction condition setting unit (19) that sets a threshold value of the correction quantification information as an extraction condition, the extraction condition being a condition for extracting the inspection image to be used for retraining from the database, and a first learning data extraction unit (21) that extracts, as learning data for retraining the region detector, the inspection image satisfying the extraction condition and the region detection result and the correction history that are associated with the inspection image.