Image Inspection Segmentation for Rule-Guided ML Evaluation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing systems face challenges in improving learning efficiency and processing speed when evaluating inspection target objects using machine learning.
Innovation Solution
An image processing device that divides original images based on features of the inspection target object into multiple regions, determines which regions can be evaluated using rule information, and sends regions that cannot be evaluated to a machine learning module for further analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all divided images are input to the machine learning module, then the machine learning module can process all cases, but the learning efficiency and processing speed deteriorate due to the large amount of data
Solution Approach 1:
The original image is divided into multiple divided images based on features of the inspection target object. This segmentation allows the system to process different regions separately, applying rule-based evaluation to simple regions and machine learning only to complex regions, thereby improving learning efficiency while maintaining evaluation completeness.
Solution Approach 2:
Different evaluation methods are applied to different divided images based on their local characteristics. Divided images with simple features use rule-based evaluation, while divided images with complex features use machine learning. This local differentiation optimizes resource allocation and improves overall processing efficiency.
2Reliability
If all divided images are input to the machine learning module, then all cases can be evaluated, but the processing speed deteriorates due to unnecessary computation
Solution Approach 1:
The image is segmented into multiple divided images, allowing the system to identify and process only the necessary portions with machine learning. This avoids unnecessary computation on regions that can be evaluated using faster rule-based methods, thereby improving processing speed while maintaining evaluation completeness.
Solution Approach 2:
Instead of applying machine learning to all divided images (excessive action), the system applies machine learning only to the necessary portion - divided images that cannot be evaluated by rule-based methods. This partial application of the more computationally intensive method optimizes processing speed.
3Measurement precision
If machine learning is used for all divided images, then accurate evaluation is achieved, but the learning efficiency deteriorates due to redundant learning data
Solution Approach 1:
The system extracts and identifies divided images that require machine learning evaluation by comparing them against rule-based evaluation criteria. Only the necessary divided images that cannot be evaluated by rules are extracted and input to the machine learning module, eliminating redundant learning data and improving learning efficiency.
Solution Approach 2:
Machine learning is applied partially only to divided images that require it, rather than excessively applying it to all divided images. This selective application reduces the amount of learning data to a necessary minimum, improving learning efficiency while maintaining evaluation accuracy.
Data Source
AI summary
Image processing device includes acquisition part, division part, determination part, and output part. Division part generates a plurality of divided images by dividing an original image based on a feature of an inspection target object shown in the original image. Determination part determines, for each of a plurality of divided images, whether or not determination as to whether an inspection target object is good or defective can be made by using rule information, and determines, for a first divided image for which determination as to whether the inspection target object is good or defective can be made, whether the inspection target object is good or defective by using the rule information. Output part outputs, to learning module capable of machine learning or obtained as a learning result, a second divided image for which determination as to whether the inspection target object is good or defective cannot be made.