Multi-Level Feature Extraction for Imbalanced Defect Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The mobile display industry faces challenges in inspecting surface defects using traditional methods, and existing AI/ML models struggle to achieve high accuracy when training datasets are highly imbalanced, leading to insufficient samples for both classes.
Innovation Solution
A system and method that uses a processor to extract features at multiple levels of abstraction, combines them, and trains a classifier, specifically a generative adversarial network, to predict faulty products from not-faulty ones, employing a pre-trained deep convolutional neural network for feature extraction and dimensionality reduction, addressing class imbalance in manufacturing defect identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional inspection mechanisms are used, then the inspection process is simple, but surface defects are harder to detect with high accuracy
Solution Approach 1:
The patent replaces traditional mechanical inspection mechanisms with an AI-based machine learning system that uses image processing and classification algorithms to detect surface defects, thereby improving detection accuracy while managing system complexity through software-based solutions
Solution Approach 2:
The patent introduces an AI classifier as an intermediary between the image capture system and the defect detection process, enabling complex pattern recognition and defect identification that bridges the gap between simple imaging and accurate defect measurement
2Reliability
If AI/ML models are trained on imbalanced datasets, then the model can handle real-world data distribution, but accuracy for both classes (good and faulty) cannot be achieved simultaneously
Solution Approach 1:
The patent applies preliminary data augmentation and resampling techniques to the training dataset before model training, creating additional synthetic samples of the minority class (faulty products) to balance the dataset and enable the model to learn both classes effectively despite limited real-world faulty samples
Solution Approach 2:
The patent modifies training parameters including class weights, loss function configuration, and sampling strategies to compensate for the imbalanced dataset, allowing the model to achieve high accuracy for both good and faulty product classes by adjusting how the model prioritizes learning from different classes
3Measurement precision
If features are extracted at multiple levels of abstraction, then the classification accuracy improves, but the computational complexity and processing time increase
Solution Approach 1:
The patent extracts features at multiple levels of abstraction from pre-trained neural network models during the offline training phase, allowing the system to learn complex defect patterns while moving the computationally intensive multi-level feature extraction to the training stage rather than real-time inspection
Solution Approach 2:
The patent uses features extracted from pre-trained models on large datasets as a form of knowledge copy, transferring learned representations to the specific defect detection task, thereby achieving high classification accuracy without requiring extensive real-time computational resources for feature extraction
Data Source
AI summary
A system and method for classifying products manufactured via a manufacturing process. A processor receives an input dataset, and extracts features of the input dataset at two or more levels of abstraction. The processor combines the extracted features and provides the combined extracted features to a classifier. The classifier is trained based on the combined extracted features for learning a pattern of not-faulty products. The trained classifier is configured to receive data for a product to be classified, to output a prediction for the product based on the received data.