Image Identification Model Selection Across Sparse Setting Space
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods using AI for lesion identification in endoscopic images require creating dedicated models for each parameter set, leading to high resource demands and reduced accuracy when conditions differ from training data.
Innovation Solution
An image processing apparatus that stores fewer machine learning models or thresholds than the total possible combinations, placed at a predetermined density in a setting space, selects and synthesizes models or thresholds based on proximity to the input image's parameters for identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If dedicated machine learning models are created for each parameter set combination, then identification accuracy is improved, but device complexity and resource demands increase
Solution Approach 1:
The patent applies universality by creating a single machine learning model that can handle multiple parameter set combinations. Instead of training separate dedicated models for each parameter set, one universal model is trained to generalize across different parameter conditions, thereby reducing the total number of models while maintaining identification accuracy.
Solution Approach 2:
The patent utilizes parameter changes by training the machine learning model on diverse training data that spans multiple parameter set combinations. The model learns to adapt its predictions based on the input parameters, allowing it to accurately identify lesions across varying parameter conditions without requiring separate models for each combination.
2Loss of energy
If the number of machine learning models is reduced, then hardware load is decreased, but identification accuracy may deteriorate
Solution Approach 1:
The patent resolves this contradiction by designing a universal machine learning model that maintains high identification accuracy while reducing the number of models. The universal model is trained on comprehensive data covering multiple parameter sets, enabling it to perform accurately across different conditions with a single model, thus lowering hardware load without sacrificing precision.
Solution Approach 2:
The patent applies copying by using synthesized training data that replicates various parameter set conditions. Instead of requiring actual separate models for each parameter combination, the system creates synthetic training examples that copy the characteristics of different parameter sets, allowing a single model to learn from diverse conditions efficiently.
3Adaptability or versatility
If machine learning models are placed densely in setting space, then coverage of parameter combinations is improved, but the number of models and resource requirements increase
Solution Approach 1:
The patent applies universality by designing a single machine learning model that can adapt to a wide range of parameter combinations. Instead of placing multiple dense models throughout the setting space, the universal model is trained on diverse data covering the entire parameter space, enabling it to generalize effectively across all conditions with just one model.
Solution Approach 2:
The patent uses partial action by training the model on a carefully selected subset of training data that represents the essential variations in parameter combinations. Rather than creating models for every possible parameter combination, the system uses representative samples that capture the key patterns, allowing the single model to generalize to unseen parameter settings effectively.
Data Source
AI summary
An image processing apparatus includes one or more processors. The one or more processors store one or more first machine learning models in a setting space including a first setting condition and a second setting condition, the one or more first machine learning models being placed at or below a predetermined density in the setting space, receives an image for identification, selects one or more second machine learning models that, in the setting space, are either within a predetermined range from the image for identification, or in order of shorter distance from the image for identification, and identifies the image for identification using the second machine learning models.