Vector Neural Network Inspection Data Quality Determination
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models, such as those using autoencoders, face challenges in accurately distinguishing between defects and inappropriate compression in inspection data, leading to potential misclassification.
Innovation Solution
A method utilizing a vector neural network with multiple vector neuron layers to classify inspection data. This involves generating training data by classifying non-defective product data into classes, executing learning using this data, preparing a known feature spectrum group, and performing quality determination by calculating similarities between inspection data and the known feature spectrum group.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a machine learning model using autoencoder is used for quality determination, then the learning can be executed using only data of non-defective products, but it may not be possible to distinguish whether the difference between inspection data and output data is due to a defect or due to inappropriate compression processing
Solution Approach 1:
The patent segments the feature extraction process by using multiple specific layers (e.g., first specific layer, second specific layer) within the vector neural network, where each layer extracts different types of features from the input data. This segmentation allows the system to capture both local and global characteristics of the inspection data, improving the ability to distinguish between defects and compression artifacts while maintaining ease of learning execution using only non-defective product data.
Solution Approach 2:
The patent transforms the quality determination problem from a direct image comparison approach to a feature spectrum analysis approach. By converting inspection data into feature spectra through multiple vector neuron layers and comparing these spectra using similarity calculations, the system operates in a transformed dimensional space that enhances the distinguishability between defects and compression artifacts, thereby improving determination accuracy.
2Reliability
If the processing of compression in an encoding unit is appropriately set, then the difference between image data can be distinguished, but inappropriate compression processing cannot be differentiated from actual defects
Solution Approach 1:
The patent changes the parameters used for quality determination by employing multiple specific layers with different feature extraction capabilities. Instead of relying on a single compression processing level, the system extracts features at multiple hierarchical levels (different layers) and combines these features to form comprehensive feature spectra. This multi-parameter approach enables the system to distinguish between compression artifacts and actual defects more reliably.
Solution Approach 2:
The patent creates a composite feature representation by combining outputs from multiple specific layers (first specific layer, second specific layer, etc.) within the vector neural network. Each layer contributes different types of features, and their combination forms a composite feature spectrum that captures both fine-grained local details and broader structural characteristics, enabling more reliable distinction between compression artifacts and defects.
3Measurement precision
If a vector neural network with multiple vector neuron layers is used, then accurate quality determination can be achieved, but the device complexity increases
Solution Approach 1:
The patent segments the vector neural network into multiple specific layers, where each layer is responsible for extracting specific types of features. This segmentation allows the complex task of quality determination to be divided into manageable sub-tasks, where each layer processes different aspects of the input data. The modular structure facilitates training and interpretation while maintaining high accuracy.
Solution Approach 2:
The patent designs the vector neural network with multiple specific layers that serve multiple functions: feature extraction, dimensionality reduction, and similarity computation. Each layer is trained to extract relevant features while the overall architecture enables both accurate quality determination and efficient similarity-based decision making, achieving multi-functionality that justifies the increased complexity.
Data Source
AI summary
A quality determination method includes: (a) generating a plurality of pieces of training data by classifying a plurality of pieces of non-defective product data into a plurality of classes; (b) executing learning of a machine learning model using the plurality of pieces of training data; (c) preparing a known feature spectrum group; and (d) executing quality determination processing of inspection data using the machine learning model and the known feature spectrum group. The (d) includes (d1) calculating a feature spectrum related to the inspection data, (d2) calculating a similarity between the feature spectrum and the known feature spectrum group, and (d3) determining the inspection data to be non-defective when the similarity is equal to or greater than a threshold value and determining the inspection data to be defective when the similarity is less than the threshold value.