Image Classification Mapping for Changing Target Classes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Training neural networks for object classification tasks is labor-intensive and requires significant effort, especially when class definitions change over time, necessitating retraining or extensive retraining.
Innovation Solution
Implement a two-stage classification process using a multi-class classifier to determine intermediate classes with confidence values, followed by a mapping to target classes, where the mapping is a simple, deterministic assignment rule or algorithm, avoiding the need for retraining when target classes change.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a neural network is trained for a specific classification task, then classification accuracy is improved, but the system requires significant retraining effort when class definitions change
Solution Approach 1:
The patent segments the classification task into two independent parts: (1) a multi-class classifier that identifies intermediate classes from image data, and (2) a mapping component that assigns target classes based on the intermediate classes. This segmentation allows the classifier to be trained once on intermediate classes while the mapping can be adjusted without retraining when target class definitions change.
Solution Approach 2:
The patent introduces intermediate classes as an intermediary layer between the image data and the target classes. The multi-class classifier processes image data to determine intermediate classes, which then serve as input for determining target classes. This intermediary layer decouples the classification process from target class definitions, enabling flexible adjustments without retraining the neural network.
2Adaptability or versatility
If class definitions are changed to adapt to new requirements, then system adaptability is improved, but retraining or extensive retraining is required
Solution Approach 1:
By separating the classification system into a trained multi-class classifier and a configurable mapping component, the patent enables independent adjustment of target class definitions. The mapping component can be modified to reflect new class definitions without affecting the trained classifier, thus improving adaptability while maintaining ease of adjustment.
Solution Approach 2:
The patent performs preliminary classification into intermediate classes that are independent of specific target class definitions. This preliminary action creates a robust foundation that can serve multiple target class schemes, allowing the system to adapt to new requirements by simply changing the mapping from intermediate to target classes rather than retraining the entire system.
3Device complexity
If a direct classification approach is used, then the system structure is simple, but the system lacks flexibility when class definitions change
Solution Approach 1:
The introduction of intermediate classes as a mediator between image data and target classes adds a layer of complexity that enables flexibility. The intermediate classes serve as a stable representation that decouples the image processing from target class definitions, allowing the system to adapt to different classification schemes without retraining while maintaining a manageable structure through the use of confidence values and mapping rules.
Data Source
Figure 1~2
Figure 3~4
AI summary
An image capture device (10) for assigning a target class to an object (14) is specified, wherein the image capture device (10) has an image sensor (20) for recording image data with the object (14) and a control and evaluation unit (22) which is designed to evaluate and classify the image data using a machine learning method, in particular a neural network, and to assign a target class to the image data. The control and evaluation unit (22) is further designed to use a multi-class classifier as a machine learning method for classifying into a plurality of intermediate classes, which classifier determines respective confidence values for assigning the image data to a respective intermediate class, and subsequently to determine the target class by applying a mapping of confidence values to target classes.