Cooking Image AI Model Updates for Recognition Accuracy Drift
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Artificial intelligence models used for cooking image analysis face challenges in maintaining recognition accuracy due to changing cooking materials and practices, and difficulty in classifying atypical or diverse food ingredients, leading to inconsistent results across different users.
Innovation Solution
An electronic apparatus that updates its artificial intelligence model based on the difference between captured image recognition accuracy and pre-stored reference accuracy, using a camera, processor, and communication interface to request and apply updates from a server when the accuracy difference exceeds a threshold, and displays a user interface for guiding the update process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the artificial intelligence model is not updated, then the device complexity and energy consumption are reduced, but the recognition accuracy drops over time due to changing cooking materials and practices
Solution Approach 1:
The patent implements a dynamic update mechanism where the AI model automatically updates when recognition accuracy drops below a threshold. The processor monitors recognition results in real-time and triggers updates only when necessary, making the system adaptable to changing cooking materials and practices without requiring manual intervention or continuous updates.
Solution Approach 2:
The system employs feedback by comparing recognition results against reference information and using this feedback to determine when updates are needed. The processor analyzes the difference between predicted and reference values, and only initiates model updates when the accuracy degradation exceeds a predetermined threshold, creating an efficient feedback loop.
2Reliability
If the artificial intelligence model is updated frequently, then the recognition accuracy is maintained, but the loss of time and energy consumption increase
Solution Approach 1:
The update frequency is dynamically adjusted based on actual recognition accuracy. Instead of fixed periodic updates, the system monitors performance and updates only when accuracy degradation is detected, minimizing unnecessary update operations and reducing time loss.
Solution Approach 2:
The system performs self-diagnosis by automatically monitoring its own recognition accuracy and triggering updates only when performance degradation is detected. This self-service mechanism eliminates the need for external monitoring and manual update scheduling, reducing time consumption.
3Reliability
If the artificial intelligence model is updated frequently, then the recognition accuracy is maintained, but the use of energy increases
Solution Approach 1:
The system dynamically adjusts energy consumption by updating the AI model only when recognition accuracy drops below a threshold. The processor monitors performance and triggers updates conditionally, reducing energy waste from unnecessary updates while maintaining accuracy when needed.
Solution Approach 2:
Energy-efficient operation is achieved through feedback mechanisms that monitor recognition accuracy and trigger updates only when performance degradation is detected. The system compares recognition results against reference information and uses this feedback to minimize unnecessary energy-consuming update operations.
4Adaptability or versatility
If the artificial intelligence model is designed to handle all food ingredients, then the adaptability is improved, but the manufacturing precision and recognition accuracy decrease for specific difficult-to-classify ingredients
Solution Approach 1:
The patent segments the recognition task by maintaining a comprehensive database of reference information for various food ingredients while using a unified AI model. The system divides the problem into general pattern recognition and specific reference matching, allowing broad adaptability while maintaining high precision for specific ingredients through comparative analysis against stored reference data.
Solution Approach 2:
Reference information stored in the database acts as an intermediary between the AI model's predictions and the final recognition result. The processor compares model predictions against this reference data, improving accuracy for difficult-to-classify ingredients by using the reference information as a mediator to verify and correct model outputs.
Data Source
AI summary
An electronic apparatus includes a camera; and at least one processor configured to: obtain, using the camera, an image captured by the camera, input the obtained image to an artificial intelligence model that is trained, obtain, using the artificial intelligence model, identification information of an object in the image and first accuracy information indicating a degree of recognition of the identification information, and update the artificial intelligence model based on a difference value between the obtained first accuracy information and second accuracy information corresponding to the identification information, wherein the second accuracy information indicates a reference degree of recognition associated with the identification information.