Zero-Shot Image Classifier for Self-Checkout Fraud Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Self-checkout machines face challenges in detecting fraudulent activities due to the need for extensive training data, which is difficult to maintain given the high variety and short life cycles of products in stores like supermarkets and convenience stores.
Innovation Solution
A data generation program that creates reference source data associating product attributes with hierarchies, reducing the data required for a zero-shot image classifier to detect fraudulent acts by using a hierarchical structure and a zero-shot image classifier that can specify product items without extensive training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional image recognition technology is used in self-checkout machines, then detection accuracy can be improved, but extensive training data is required which is difficult to maintain given high product variety and short life cycles
Solution Approach 1:
The patent segments the product attribute space into hierarchical categories (e.g., product type, brand, specifications) rather than treating each product individually. This segmentation allows the system to generalize across product variations without requiring separate training data for each product, thereby maintaining detection accuracy while reducing data maintenance complexity.
Solution Approach 2:
The patent creates a universal attribute-based representation that can apply to any product regardless of its specific type or lifecycle. By focusing on inherent attributes rather than product-specific features, the system achieves multi-functionality across diverse product categories, eliminating the need for extensive product-specific training data while maintaining fraud detection capabilities.
2Measurement precision
If product data is stored in detail for each product variant, then detection precision is improved, but data storage and processing costs increase significantly
Solution Approach 1:
The patent extracts only the essential attribute features from complete product data while discarding redundant information. By taking out only the critical attributes needed for fraud detection (such as product category, key specifications, and identifying characteristics), the system maintains identification precision while significantly reducing the quantity of stored data.
Solution Approach 2:
The patent transforms the data structure from storing detailed information about each individual product variant to storing attribute-based representations across multiple dimensions. This dimensional transformation allows the system to represent diverse products using a compact attribute framework, reducing storage requirements while preserving the ability to precisely identify and differentiate products through attribute comparison.
Data Source
AI summary
A non-transitory computer-readable storage medium storing a data generation program that causes at least one computer to execute a process, the process includes acquiring product data; generating reference source data in which attributes of products are associated with each of a plurality of hierarchies based on a variance relationship of attributes of products included in the acquired product data; and setting the generated reference source data as data to be referred to by a zero-shot image classifier.