Neural Network Alignment for Product Defect Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current auto visual inspection technologies face inefficiencies in aligning product images due to variability such as rotation, scaling, and shearing, leading to time-consuming defect detection processes in manufacturing.
Innovation Solution
A computer-implemented method and system that generates geometric training parameters to transform template images, using a neural network to align product images under inspection with template images, significantly reducing processing time by leveraging self-learning and random transformation parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional alignment methods are used to account for rotation, scaling, and shearing variability in product images, then alignment accuracy is improved, but processing time increases significantly
Solution Approach 1:
The patent applies preliminary action by pre-training a neural network model with geometric transformation parameters (rotation, scaling, shearing) during the data preparation phase. This allows the model to learn alignment patterns in advance, so that during actual defect detection, the pre-trained model can quickly align images without performing computationally intensive geometric transformations in real-time, thus resolving the contradiction between alignment accuracy and processing time
Solution Approach 2:
The patent replaces traditional mechanical alignment methods (which involve explicit geometric transformations and parameter adjustments) with a neural network-based learning system. The neural network learns to perform alignment through training on transformed template images, substituting the mechanical transformation process with an intelligent model that can rapidly predict alignment parameters, thereby reducing processing time while maintaining accuracy
2Productivity
If geometric transformation parameters are generated and applied to template images for training, then alignment speed is improved, but data preparation complexity increases
Solution Approach 1:
The patent applies parameter changes by systematically varying geometric transformation parameters (rotation angles, scaling factors, shearing values) during the training data generation phase. These parameter changes create diverse transformed template images that teach the neural network to handle various alignment scenarios. The controlled manipulation of transformation parameters enables the model to learn robust alignment capabilities, achieving high alignment speed during inference while the complexity is confined to the one-time training phase
Data Source
AI summary
Embodiments of the present invention facilitate product defect detection. A computer-implemented method comprises: receiving, by a device operatively coupled to one or more processors, a template image of a normal product; generating, by the device, one or more geometric training parameters for transforming the template image; and transforming, by the device, the template image using the one or more geometric training parameters to generate a transformed image for training a data model, wherein the trained data model being used for aligning the template image and an image under inspection of a product.