Learning Device for Adaptive Appearance Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing appearance inspection devices perform uniform inspections across multiple directions, which are inadequate for varying anomalies on different surfaces or parts of a product, as they do not account for unique features of each plane or portion.
Innovation Solution
A learning device that acquires time-series images of a target object and simultaneously trains a group discrimination model and recognition models to classify images into groups, allowing for tailored abnormality determination by integrating results from multiple recognition models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If the same inspection is performed in three directions with respect to an image of an object to be inspected, then comprehensive coverage of the object is achieved, but the inspection method cannot adapt to varying anomalies on different surfaces or parts of the product
Solution Approach 1:
The patent segments the inspection process by dividing images into multiple groups based on features such as defect presence, product type, or inspection priority. Different recognition models are trained and applied to different groups, allowing each model to specialize in specific anomaly types or product portions rather than using a single uniform inspection approach for all images.
Solution Approach 2:
The patent applies local quality by assigning different recognition models to different image groups based on their specific characteristics. Each recognition model is tailored to handle particular types of anomalies or product regions, enabling the inspection system to adapt its analysis method to the local features of each image group rather than applying a generic inspection method uniformly.
2Measurement precision
If multiple recognition models are trained separately for different image groups, then accurate abnormality determination for each group is achieved, but the training process becomes time-consuming and inefficient
Solution Approach 1:
The patent merges the training processes of multiple recognition models by enabling simultaneous training of all models in parallel. The learning device trains multiple recognition models concurrently on different image groups rather than sequentially, significantly reducing the total training time while maintaining the specialized accuracy benefits of multiple models.
Solution Approach 2:
The patent implements preliminary action by pre-dividing images into groups based on their features before training begins. This pre-segmentation allows the simultaneous training process to efficiently allocate computational resources to each group's specialized model without interference, preparing the data structure in advance to enable parallel processing during the training phase.
Data Source
AI summary
In a learning device, an acquisition means acquires captured images in a time series which capture a target object. Next, a learning means simultaneously trains a group discrimination model for discriminating a plurality of groups from the captured images based on features in each image and a plurality of recognition models each for recognizing captured images belonging to a corresponding group.