Autoencoder Feature Dimension Tuning for Image Defect Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting product defects in images using autoencoders are inefficient due to the lack of optimal dimensional settings for underlying features, leading to suboptimal defect detection efficiency.
Innovation Solution
The method involves selecting and optimizing the underlying feature dimension of an autoencoder using a t-distributed stochastic neighbor embedding (t-SNE) algorithm, which reduces feature dimensions and computes scores to determine optimal settings for effective defect detection, involving steps like training the autoencoder with sample data, dimension reduction, and calculating reconstruction errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the underlying feature dimension of the autoencoder is not optimally set, then the defect detection efficiency is low, but increasing the dimension may increase computational complexity
Solution Approach 1:
The patent applies parameter changes by systematically varying the underlying feature dimension parameter of the autoencoder to identify the optimal value. The method trains multiple autoencoders with different feature dimensions and evaluates their performance using defect detection metrics, thereby selecting the dimension that maximizes detection efficiency while managing computational complexity.
Solution Approach 2:
The patent employs preliminary action by conducting a preliminary experiment phase where different feature dimensions are tested before final model deployment. This preliminary exploration of parameter space allows the system to pre-identify the optimal dimension, avoiding the need to test all possible dimensions during actual defect detection operations.
2Measurement precision
If the underlying feature dimension is increased to improve discernibility, then defect detection accuracy improves, but training time and computational resources increase
Solution Approach 1:
The patent systematically changes the feature dimension parameter to find the optimal balance between accuracy and training time. By evaluating multiple dimension values and selecting the one that achieves sufficient defect detection accuracy with minimal training time, the method resolves the contradiction between precision and time consumption.
Solution Approach 2:
The patent applies partial action by not using the maximum possible feature dimension but rather a sufficient dimension that achieves the required detection accuracy. This approach avoids the excessive computational cost of using overly large dimensions while still maintaining adequate defect detection performance.
Data Source
AI summary
A method of detecting defects revealed in images of products obtains sample image training data. An underlying feature dimension of an autoencoder is selected and a score is obtained. By comparing the score with a standard score, an optimal underlying feature dimension is confirmed. A test image is inputted into the autoencoder with the optimal underlying feature dimension to obtain a reconstruction image. A reconstruction error between the test image and the reconstruction image is computed. By comparing the reconstruction error with the predefined threshold a result of analysis of the test image is outputted. An image defect detection apparatus and a computer readable storage medium applying the method are also provided.