Multi-Neural Network System for Gel Card Reaction Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated recognition systems for blood agglutination reactions in gel cards face difficulties in recognizing atypical reactions and require frequent recalibration due to changes in camera type, transparency, or light conditions, leading to inefficiencies.
Innovation Solution
A multi-neural network system is employed, where a first neural network provides probabilities for initial reaction categories, and a second neural network analyzes a predetermined portion of the image if the initial category's probability exceeds a threshold, with a screening stage determining image suitability for analysis, reducing the need for frequent recalibration.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If image-processing based recognition apparatuses are used to automatically determine reactions, then efficiency is improved, but the apparatus requires frequent recalibration when camera type, gel card transparency, or light conditions change
Solution Approach 1:
The patent replaces traditional image-processing software with a neural network-based system. The neural network learns to recognize agglutination reactions directly from raw images, eliminating the need for manual image processing and recalibration. This substitution of mechanical/image-processing systems with an intelligent neural network system resolves the contradiction by maintaining high efficiency while eliminating frequent recalibration needs.
Solution Approach 2:
The patent changes the fundamental parameter of the recognition system from traditional image processing algorithms to neural network models. This parameter change allows the system to adapt to different camera types, gel card transparencies, and lighting conditions without recalibration, as the neural network learns these variations during training. This resolves the contradiction by enabling the system to maintain efficiency across varying conditions without frequent recalibration.
2Extent of automation
If traditional image-processing software is used to recognize reactions, then automation is achieved, but atypical reactions cannot be recognized accurately
Solution Approach 1:
The patent substitutes traditional image-processing software with a neural network system that automatically recognizes reactions including atypical ones. The neural network's ability to learn from diverse training data enables it to accurately identify atypical reactions while maintaining automation. This substitution resolves the contradiction by achieving both automation and high recognition accuracy for atypical reactions.
Solution Approach 2:
The patent implements a feedback mechanism where the neural network continuously learns from training data and can be fine-tuned to improve recognition of atypical reactions. The system's ability to process and learn from diverse reaction patterns enables accurate recognition of atypical cases while maintaining automated operation. This feedback mechanism resolves the contradiction by enabling the system to adapt to and accurately recognize atypical reactions through automated learning processes.
3Device complexity
If a single neural network is used for all reaction categories, then device complexity is reduced, but the system cannot handle different reaction types with varying levels of detail
Solution Approach 1:
The patent segments the neural network into multiple specialized networks, each dedicated to recognizing specific reaction categories. This segmentation allows each network to be optimized for its specific reaction type, improving classification accuracy while keeping the overall system architecture manageable. The segmentation principle resolves the contradiction by enabling specialized processing for different reaction types without creating an overly complex unified system.
Solution Approach 2:
The patent creates a multi-functional neural network system where multiple specialized networks work together to handle different reaction categories. Each network serves a specific function (recognizing different reaction types), but they collectively provide universal coverage for all possible reactions. This multi-functionality approach resolves the contradiction by achieving high accuracy for diverse reaction types while maintaining reasonable system complexity through modular organization.
Data Source
AI summary
An apparatus for classifying a picture of a reaction of reactants in a predetermined container; the apparatus comprising a first neural network arranged for receiving an input picture of a reaction of reactants in a predetermined container and for providing, for each reaction category of a first plurality of reaction categories, a probability that the input picture shows a reaction that belongs to said reaction category; a second neural network arranged for, if the first neural response provides a highest probability that the input picture shows a reaction that belongs to a predetermined reaction category, receiving a predetermined portion of the input picture and providing, for each reaction category of a second plurality of reaction categories, a probability that said predetermined portion of the input picture shows a reaction that belongs to said reaction category.