Image Classification Mapping for Changing Target Classes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge of adapting neural networks for object classification tasks when class definitions change over time involves significant retraining efforts, particularly in applications like logistics and automation.
Innovation Solution
A two-stage classification process using a multiclass classifier to determine intermediate classes with confidence values, followed by a mapping to target classes through a simple, deterministic association rule, avoiding the need for complex retraining.
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 adaptation to new class definitions requires significant retraining effort and time
Solution Approach 1:
The classification task is segmented into two independent stages: (1) training a multiclass classifier for intermediate classes, and (2) creating a mapping from intermediate classes to target classes. This segmentation allows the first stage to be trained once and reused, while only the second stage needs adaptation when class definitions change.
Solution Approach 2:
Intermediate classes serve as an intermediary layer between the neural network's native classification output and the application-specific target classes. This intermediary enables decoupling the complex neural network training from the flexible target class definitions, allowing rapid adaptation through mapping adjustments rather than full retraining.
2Reliability
If a multiclass classifier is trained extensively for accurate classification, then classification reliability is improved, but the complexity of the system increases
Solution Approach 1:
The system is segmented into a trained multiclass classifier component and a separate mapping component. This segmentation allows the complex training to be confined to the first component, while the second component remains simple and interpretable, reducing overall system complexity.
Solution Approach 2:
Instead of directly training the neural network for each specific target class scenario, a general multiclass classifier is trained once to produce intermediate class predictions, which are then copied and mapped to various target class configurations. This copying approach avoids repeating complex training for each scenario.
3Device complexity
If direct classification to target classes is performed, then the classification process is simplified, but adaptability to changing class definitions deteriorates
Solution Approach 1:
The mapping from intermediate classes to target classes is designed to be dynamic and configurable. When new target classes or definitions are introduced, the mapping can be adjusted without retraining the multiclass classifier, enabling the system to adapt to changing requirements while maintaining a simple core classification process.
Solution Approach 2:
The intermediate classes act as a flexible intermediary layer that decouples the stable multiclass classifier from the potentially changing target class definitions. This intermediary enables easy adaptation by modifying only the mapping relationships, not the core classification process.
Data Source
AI summary
An image capturing device (10) for associating a target class with an object (14) is provided, wherein the image capturing device (10) has an image sensor (20) for recording image data having the object (14) and a control and evaluation unit (22) that is configured to evaluate and classify the image data using a method of machine learning, in particular a neural network, and to associate a target class with the image data. In this respect, the control and evaluation unit (22) is further configured to use as a method of machine learning a multiclass classifier for the classification into a plurality of intermediate classes that determines respective confidence values for the association of the image data with a respective intermediate class and subsequently to determine the target class by applying a map of confidence values in target classes.