Retail Product Image Labeling With Selective Model Retraining
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Manual inspection of product storage facilities is time-consuming and costly, as workers need to visually check inventory levels across numerous shelves and pallets, diverting resources from other tasks.
Innovation Solution
Implementing a system with a movable image capture device equipped with machine learning algorithms to automatically capture and process images, cluster similar products, and retrain the model using selected samples to improve recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual inspection is used to identify products on shelves, then workers can visually check inventory levels, but the process becomes time-consuming and increases operational costs
Solution Approach 1:
The patent replaces the mechanical/visual inspection system with an automated image processing system. Image capture devices take photographs of shelves, and machine learning models automatically identify products and their locations, eliminating the need for manual visual inspection while maintaining or improving identification accuracy.
Solution Approach 2:
The system enables self-service inventory monitoring where the machine learning model autonomously processes images, identifies products, and updates inventory records without human intervention. The model continuously improves through automated retraining on new data, making the system self-enhancing over time.
2Reliability
If workers manually inspect all product storage areas, then complete inventory coverage is achieved, but resource allocation efficiency decreases
Solution Approach 1:
The patent replaces human workers with an automated system consisting of image capture devices and machine learning models. This substitution maintains complete inventory monitoring coverage while freeing workers to perform other value-added tasks, thereby improving overall productivity.
Solution Approach 2:
The system introduces an intermediary layer between physical inventory and digital records. Image capture devices and machine learning models act as intermediaries that automatically translate visual information into structured inventory data, ensuring completeness without requiring direct human involvement in each inspection.
3Measurement precision
If machine learning models are trained on all available images, then recognition accuracy improves, but processing time and computational resources increase
Solution Approach 1:
The patent extracts only the most valuable training samples from the full image dataset. The system identifies images that are most likely to improve model performance and uses only those for retraining, rather than processing all available images. This extraction approach maintains accuracy improvement while significantly reducing training time and computational resources.
Solution Approach 2:
The system performs partial training by selecting a subset of training images rather than using the complete dataset. This partial action is sufficient to maintain and improve model accuracy while avoiding the excessive computational burden of training on all available images.
Data Source
AI summary
In some embodiments, apparatuses and methods are provided herein useful to processing captured images. In some embodiments, there is provided a system for processing captured images of objects including a memory and a control circuit executing a trained machine learning model. The memory may be configured to store a plurality of images comprising first images and second images. The control circuit may be configured to: allocate each of the first images into one of a plurality of datasets; cluster each image in the dataset into one of a plurality of groups; select a sample from at least one of the plurality of groups; cluster each of the second images into one of dominant product identifier group and a non-dominant product identifier group; select a sample from the dominant product identifier group and a sample from the non-dominant product identifier group; and output the selected sample.