Image Classification Model Global Class Feature Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image classification methods face challenges in accurately classifying images when training data is insufficient, particularly for novel classes with few training samples, as they often overfit to base classes with abundant data.
Innovation Solution
The proposed method involves an image classification model that uses a global class feature trained on both base and novel classes, updated through episodic training, to avoid overfitting. This model extracts feature vectors, determines local and global class features, and adjusts the global class feature based on classification and registration errors to improve identification of novel class images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning algorithms are used for image classification, then classification accuracy is improved, but the method highly relies on large amounts of tagged training data which is difficult to obtain
Solution Approach 1:
The training set is segmented into base classes (with abundant data) and novel classes (with scarce data). The model is trained separately on base classes first, then adapted to novel classes using few-shot learning techniques, allowing effective classification without requiring large amounts of data for each class
Solution Approach 2:
The model performs preliminary training on base classes with abundant training data before being applied to novel classes. This preliminary action pre-learns general features and patterns that can be transferred to novel classes, reducing the dependency on large amounts of novel class training data
2Adaptability or versatility
If the model is trained on base class images, then the model learns general features, but the model overfits to base class and fails to accurately identify novel class images
Solution Approach 1:
The model uses feedback from both base class and novel class training images to continuously update and refine the global class feature. This feedback mechanism ensures the model learns general features from base classes while also adapting to novel classes, preventing overfitting to any single class
Solution Approach 2:
The global class feature serves multiple functions: it captures general visual features from base classes while also representing novel classes. This universal feature representation allows the model to generalize across different class types without overfitting to specific base class characteristics
Data Source
AI summary
An image classification method and apparatus, and an image classification model training method and apparatus are provided, which are related to an image recognition technology in the artificial intelligence field and more specifically, to the computer vision field. The method includes: obtaining a to-be-processed image; and classifying the to-be-processed image based on a preset global class feature, to obtain a classification result of the to-be-processed image. The preset global class feature includes a plurality of class features obtained through training based on a plurality of training images in a training set. The plurality of class features in the preset global class feature are used to indicate visual features of all classes in the training set.