Neural Network Segmentation for Automated Inventory Catalog Property Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current delivery systems lack the ability to automatically update item property information, such as organic or vegan status, in inventory catalogs due to the difficulty in recognizing and translating visual symbols on item packaging, which is crucial for consumer decision-making.
Innovation Solution
Utilizing machine-learned segmentation and classification neural networks to identify and classify item labels from images, enabling automated updating of item properties in inventory databases without manual intervention.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If manual input is used to add item properties to the inventory catalog, then item property information can be added, but the process is difficult and time-consuming
Solution Approach 1:
The patent replaces the manual mechanical process of visually observing item labels and translating them to item properties with an automated optical system. The system captures images of item labels using an image capture unit and uses machine learning models (segmentation model and classification model) to automatically identify and classify item properties, eliminating the need for manual human intervention in the information extraction process.
2Productivity
If automated systems are used to recognize item labels, then productivity improves, but the system complexity increases
Solution Approach 1:
The patent divides the complex task of item label recognition into two distinct sequential stages: (1) a segmentation model that identifies and extracts regions containing item labels from the captured image, and (2) a classification model that analyzes the extracted label regions to determine specific item properties. This segmentation of the recognition process reduces overall system complexity by breaking down the monolithic task into manageable, specialized components.
Solution Approach 2:
The patent introduces an intermediate processing stage where the segmentation model acts as a mediator between the raw image capture and the final classification. The segmentation model extracts and prepares label regions as intermediate data structures, which are then passed to the classification model. This intermediary step simplifies the classification task by providing pre-processed, focused input data.
Data Source
Figure 1
Figure 2
Figure 3A~3B
AI summary
A method for populating an inventory catalog includes receiving an image showing an item in the inventory catalog and comprising a plurality of pixels. A machine learned segmentation neural network is retrieved to determine location of pixels in an image that are associated with an image label and the property. The method determines a subset of pixels associated with the item label in the received image and identifies locations of the subset of pixels of the received image, and extracts the subset of pixels from the received image. The method retrieves a machine learned classifier to determine whether an image shows the item label. The method determines, using the machine learned classifier, that the extracted subset of pixels shows the item label. The method updates the inventory catalog for the item to indicate that the item has the property associated with the item label.