Characteristic-Based Classifier Retrieval for Product Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current state-of-the-art technologies, such as large-scale deep neural networks, face limitations in recognizing a large number of products due to the classifier-capacity problem and fine-grained recognition challenges, where products with similar packaging differ only by small graphic elements, leading to increased errors in product identification.
Innovation Solution
The system employs a characteristic-based classifier training and retrieval method, utilizing multiple classifiers trained on similar product characteristics, with imaging modules and a central processing module to extract and match product features from a database, enhancing recognition accuracy by selecting relevant classifiers for each product.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single classifier is used to recognize products, then the device complexity is low, but the recognition accuracy deteriorates when the number of products increases
Solution Approach 1:
The patent divides the product recognition task into multiple independent classifiers, each trained on a specific subset of products or product characteristics. Instead of using one large classifier for all products, the system segments the classifier population into multiple specialized classifiers that work together to recognize the full product catalog, thereby maintaining high accuracy while managing complexity
Solution Approach 2:
The patent creates a universal classifier system where multiple classifiers serve different product categories or characteristics but can collectively recognize any product in the catalog. Each classifier is designed to handle specific product types or features, yet the overall system maintains universality by combining these specialized classifiers to achieve broad product recognition capability
2Adaptability or versatility
If the number of products to be recognized is increased, then the system versatility improves, but the recognition accuracy deteriorates due to the classifier-capacity problem
Solution Approach 1:
The patent segments the large product catalog into multiple smaller subsets, each handled by a dedicated classifier. This segmentation allows each classifier to maintain high accuracy on its specific subset while the collective system achieves high versatility across the entire product catalog, effectively resolving the classifier-capacity problem
Solution Approach 2:
The patent introduces a new dimension to product recognition by training classifiers on product characteristics and features rather than just product identifiers. This dimensional shift allows the system to recognize products based on their attributes, enabling high versatility in recognizing new products while maintaining accuracy through characteristic-based classification
3Measurement precision
If products with similar packaging are distinguished by small graphic elements, then the measurement precision requirement increases, but the difficulty of detecting and measuring increases
Solution Approach 1:
The patent applies local quality by training classifiers to focus on specific local features or graphic elements on product packaging rather than the entire package. Each classifier is specialized to detect and distinguish products based on particular local characteristics such as logos, labels, or specific graphic elements, thereby achieving high precision in fine-grained product distinction while managing the complexity of detecting small details
Data Source
AI summary
The disclosure relates to systems and methods for real-time detection of a very large number of items in a given constrained volume. Specifically, the disclosure relates to systems and methods for retrieving an optimized set of classifiers from a self-updating classifiers' database, configured to selectively and specifically identify products inserted into a cart in real time, from a database comprising a large number of stock-keeping items, whereby the inserted items' captured images serve simultaneously as training dataset, validation dataset and test dataset for the recognition/identification/re-identification of the product.