Image Inspection Apparatus Using Segmented Normal and Deep Learning Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image inspection technologies face challenges in achieving high inspection accuracy and reducing processing time, particularly when dealing with unclear characteristic amounts in images of metal components or those with color unevenness, and deep learning methods are prone to unstable behavior with unknown data and longer processing times.
Innovation Solution
An image inspection apparatus that combines normal inspection processing and deep learning processing, where normal inspection is used for clear characteristic amounts and deep learning is applied for unclear cases, utilizing a neural network with three or more layers to classify images as non-defective or defective products, thereby improving throughput and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning processing is used to improve inspection accuracy for unclear characteristic amounts, then inspection accuracy is improved, but processing time increases
Solution Approach 1:
The inspection system segments the inspection process into two distinct paths: normal inspection processing for clear characteristic amounts and deep learning processing for unclear characteristic amounts. This segmentation allows the system to apply the appropriate processing method to each case, avoiding unnecessary deep learning processing for clear cases and thereby reducing overall processing time while maintaining high inspection accuracy for difficult cases.
2Adaptability or versatility
If deep learning processing is used to handle unclear characteristic amounts, then inspection capability is improved, but behavior becomes unstable with unknown data
Solution Approach 1:
The system introduces an intermediary mechanism - a determination unit that evaluates whether the characteristic amount is clear or unclear before selecting the processing method. This intermediary acts as a gatekeeper, directing clear cases to normal inspection and unclear cases to deep learning processing, thereby preventing deep learning from processing unknown data inappropriately and maintaining determination stability.
3Productivity
If normal inspection is used for clear characteristic amounts, then processing speed is maintained, but inspection accuracy decreases for unclear cases
Solution Approach 1:
The system dynamically adjusts the inspection method based on the clarity of the characteristic amount. The determination unit continuously evaluates each inspection case and switches between normal inspection and deep learning processing accordingly. This dynamic adaptation allows the system to maintain high processing speed for clear cases while ensuring high inspection accuracy for unclear cases.
Data Source
AI summary
When a normal inspection and an inspection through deep learning processing is applicable, high inspection accuracy is obtained while reducing a processing time. The normal inspection processing is applied to a newly acquired inspection target image, the non-defective product determination or the defective product determination is confirmed for the inspection target image having the characteristic amount with which the non-defective product determination or the defective product determination is executable based on the characteristic amount within the inspection target image and the threshold for confirming the non-defective product determination or the threshold for confirming the defective product determination. The deep learning processing is applied to the inspection target image having the characteristic amount with which the non-defective product determination or the defective product determination is not confirmable, and the non-defective product determination or the defective product determination is executed.