Voice-Guided Object Recognition Model Updates for New Categories
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing deep learning-based object recognition models can only recognize objects from categories seen during training, failing to identify new or unseen categories.
Innovation Solution
Update the object recognition model using a target image and voice information to incorporate new features and labels, ensuring high similarity and confidence in the indicated category, thereby expanding recognition capabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the object recognition model is trained using a large quantity of object images of known categories, then the model can learn unique features of different categories and achieve accurate recognition for trained categories, but the model cannot recognize new or unseen categories that were not present during training
Solution Approach 1:
The system enables users to manually add new object categories and training images without requiring retraining of the entire model. Users can simply provide images and labels, and the system automatically updates the model's feature library to include these new categories, allowing the model to serve itself by adapting to new objects without external intervention
Solution Approach 2:
The patent changes the parameter of the model's feature library by adding new feature-image label mappings. Instead of changing the fundamental model architecture or retraining all parameters, the system updates the feature library with new entries, allowing the model to adapt to new categories by modifying its parameter storage rather than its core structure
2Adaptability or versatility
If the object recognition model is updated by adding new categories through manual training, then the model can recognize more object categories, but the process requires complex retraining procedures and large amounts of labeled data
Solution Approach 1:
The patent segments the model into two functional parts: a pre-trained feature extraction component and a configurable feature library. The feature library acts as a separate, easily updatable component that stores image-label mappings. This segmentation allows new categories to be added by simply updating the library rather than retraining the entire model, reducing complexity
Solution Approach 2:
The system performs preliminary action by pre-training the feature extraction capabilities and storing them in the model. This preliminary training handles the complex pattern recognition tasks, while the feature library simply stores the mappings between extracted features and labels. When new categories are needed, only the library needs updating, not the pre-trained features
3Reliability
If the model uses deep learning-based object recognition, then it can achieve high recognition accuracy for known categories, but it lacks the ability to adapt to new categories without retraining
Solution Approach 1:
The feature library acts as an intermediary between the pre-trained deep learning model and the new object categories. The deep learning model extracts features from images, the feature library stores the correspondence between these features and labels, and this intermediary structure allows the system to adapt to new categories by simply updating the library rather than retraining the deep learning components
Data Source
AI summary
A method and apparatus for updating an object recognition model in the field of artificial intelligence are disclosed. According to the method, a target image and first voice information of a user are obtained. The first voice information indicates a first category of a target object in the target image. A feature library of a first object recognition model is updated based on the target image and the first voice information. The updated first object recognition model includes a feature of the target object and a first label indicating the first category, and the feature of the target object corresponds to the first label. A recognition rate of an object recognition model can be improved more easily according to the technical solution provided in this application.