Dynamic Centroid Classification for Flexible Image Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional deep learning image recognition models are inflexible and resource-intensive, making it difficult to add or modify categories without re-training the entire model, while search-based models consume excessive memory and computation resources.
Innovation Solution
A method that appends feature extraction layers to generate centroids for clustering feature vectors, allowing for a dynamic classification layer that can be updated without re-training, balancing model size and recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a traditional deep learning image recognition model with a fixed classification layer is used, then the model structure is simple and computational efficiency is optimized, but the model cannot easily add or modify categories without re-training the entire model
Solution Approach 1:
The patent segments the classification module into a fixed pre-trained classification layer and a dynamic centroid database. The centroid database can be independently updated by adding, removing, or modifying centroids without re-training the entire model, thus enabling flexible category modifications while preserving the pre-trained features.
Solution Approach 2:
The patent introduces a dynamic centroid database that can be updated in real-time without re-training. The classification module dynamically adapts to new categories by adding new centroids to the database, making the system flexible and adaptable to changing classification requirements.
2Adaptability or versatility
If a search-based recognition model with a database of indexed feature vectors is used, then the model is flexible with category modifications, but the database size is large consuming excessive memory and computation resources
Solution Approach 1:
The patent extracts only the essential classification information (centroids) from the full feature database. Instead of storing and searching all training images, the system stores only the compressed centroid representations, dramatically reducing memory usage while maintaining classification flexibility.
Solution Approach 2:
The patent creates a simplified copy of the classification knowledge in the form of centroids, which are compressed representations of category features. These centroids serve as a lightweight alternative to the full feature database, enabling flexible category modifications with minimal resource consumption.
3Ease of manufacture
If a search-based recognition model is used, then the model can be fine-tuned without re-training neural network layers, but the large database size makes it difficult to deploy on resource-constrained systems like mobile devices
Solution Approach 1:
The patent extracts only the necessary classification centroids from the full feature database, creating a lightweight model that can be easily deployed on mobile devices. The extracted centroids maintain the fine-tuning capability while reducing the model size to be suitable for resource-constrained environments.
Data Source
AI summary
A method of generating an image recognition model for recognising an input image and a system thereof are provided. The method includes appending at least one feature extraction layer to the image recognition model, extracting a plurality of feature vectors from a set of predetermined images, grouping the plurality of feature vectors into a plurality of categories, clustering the plurality of feature vectors of each of the plurality of categories into at least one cluster, determining at least one centroid for each of the at least one cluster, such that each of the at least one cluster comprises at least one centroid, such that each of the at least one centroid is represented by a feature vector, generating a classification layer based on the feature vector of the at least one centroid of the plurality of categories, and appending the classification layer to the image recognition model. In addition, a method of classifying an input image and a system thereof are provided.