Multi-Hierarchy Classification for Product Contour Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current inventory management systems face challenges in accurately counting and locating products on shelves due to issues like items being placed on wrong shelves or improperly oriented, which existing automation technologies such as RFID and scale sensors cannot effectively address.
Innovation Solution
A computer-implemented method using a multi-hierarchy classification system that employs machine learning algorithms to identify product contours and backgrounds in images, distinguishing between products and unknown items by training smart contour models and shape-specific classification models, and utilizing an ultra-large scale multi-hierarchy classification system for accurate product identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If scale sensors are used to detect inventory levels, then automated counting is improved, but the system cannot detect improper orientation or wrong shelf placement
Solution Approach 1:
The patent divides the inventory detection task into multiple specialized components: weight sensors for quantity detection, computer vision systems for orientation and location verification, and RFID readers for product identification. Each component handles a specific aspect of inventory monitoring, allowing the system to gather comprehensive information that no single sensor type could provide alone.
2Extent of automation
If RFID tagging is used for automation, then inventory tracking is improved, but the system cannot verify proper product orientation or placement
Solution Approach 1:
The patent introduces computer vision cameras as an intermediary component that bridges the gap between RFID tracking and physical verification. The vision system captures images of products on shelves, and image processing algorithms analyze these visuals to determine product orientation, placement accuracy, and orientation compliance, supplementing the automated tracking data from RFID tags.
3Measurement precision
If manual inventory checking is used, then product orientation and placement can be verified, but productivity and automation are reduced
Solution Approach 1:
The patent implements a self-monitoring inventory system where weight sensors automatically detect product placement and removal, computer vision systems autonomously analyze product orientation and placement compliance, and the system generates automated alerts for exceptions. This eliminates the need for manual inventory checking while maintaining high measurement precision, as the system continuously monitors and verifies product status without human intervention.
4Device complexity
If a single classification model is used for product identification, then system complexity is reduced, but accuracy in distinguishing similar products decreases
Solution Approach 1:
The patent implements a hierarchical classification system that segments product identification into multiple levels: first-classification models that categorize broad product types, and second-classification models that identify specific product variants. This multi-level approach improves identification accuracy for similar products while managing system complexity through structured organization of classification tasks.
Solution Approach 2:
The patent enhances product differentiation by analyzing multiple dimensions of product characteristics simultaneously: visual features from images, weight measurements, RFID data, and spatial position information. By considering products from multiple dimensional aspects rather than relying on a single classification model, the system achieves higher identification accuracy for similar products.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
In an example, a first machine learning algorithm is used to train a smart contour model to identify contours of product shapes in input images and to identify backgrounds in the input images. A second machine learning algorithm is used to train a plurality of shape-specific classification models to output identifications of products in input images. A candidate image of one or more products is obtained. The candidate image is passed to the smart contour model, obtaining output of one or more tags identifying product contours in the candidate image. The candidate image and the one or more tags are passed to an ultra-large scale multi-hierarchy classification system to identify one or more classification models for one or more individual product shapes in the candidate image. The one or more classification models are used to distinguish between one or more products and one or more unknown products in the image.