Autoencoder Image Defect Detection Using Gaussian Mixture Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image defect detection methods require a large number of defect sample images, leading to reduced accuracy due to the scarcity of such images.
Innovation Solution
An image defect detection method utilizing an autoencoder to process flawless sample images, generate hidden and reconstructed vectors, calculate reconstruction errors, and train a Gaussian mixture model to establish a defect detection model, improving accuracy by predicting defect samples.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large number of defect sample images are used for training, then the detection model can be trained, but the scarcity of defect samples limits the accuracy of defect detection
Solution Approach 1:
The patent applies preliminary action by using flawless sample images to pre-train the autoencoder model before actual defect detection. The autoencoder learns normal product features in advance through training on flawless samples, enabling it to reconstruct normal patterns accurately. When defect detection is performed, the model compares reconstructed images with original images, and discrepancies indicate defects. This preliminary training on abundant flawless data compensates for the scarcity of defect samples.
Solution Approach 2:
The patent uses copying by generating reconstructed images through the autoencoder model. The autoencoder creates copies (reconstructions) of the input flawless sample images, and the difference between original and reconstructed images serves as the basis for defect detection. This copying mechanism allows the system to simulate defect detection scenarios using only flawless samples, effectively overcoming the lack of actual defect images for training.
2Measurement precision
If traditional defect detection methods are used, then the process is simple, but the accuracy is low due to insufficient defect sample data
Solution Approach 1:
The patent replaces traditional mechanical defect detection methods with an intelligent autoencoder-based system. Instead of relying on simple threshold comparisons or rule-based detection, the system uses a deep learning autoencoder model that automatically learns feature representations from flawless sample images. This substitution enables the system to achieve high detection accuracy by capturing complex patterns and subtle defects that traditional methods would miss, despite the increased model complexity.
Data Source
AI summary
An image defect detection method is used in an electronic device. The electronic device determines a training image feature set, and trains a Gaussian mixture model by using the feature set to obtain an image defect detection model and a reference error value. An image for analysis is input into the autoencoder to obtain a second implicit vector and a second reconstructed image, and to calculate a second reconstruction error. The electronic device obtains a test image feature of the image for analysis according to the second reconstruction error and the second implicit vector, and inputs the test image feature into the image defect detection model to obtain a prediction score. The image for analysis is determined to reveal a defect when the prediction score is less than or equal to the reference error value.