Researchers at Pitt are demonstrating the effectiveness of machine learning techniques in enhancing the capabilities of barcode scanning systems, particularly for UPC-A barcodes. The improved accuracy and robustness of the machine learning-based approaches, compared to conventional methods, suggest that this technology holds great potential for driving innovation and improving processing times in various industrial applications where fast and reliable barcode scanning is crucial.
Description
The researchers propose a machine learning-based approach for decoding UPC-A barcodes, which outperforms conventional barcode scanner algorithms in terms of accuracy. Conventional barcode scanning technology relies on signal processing and edge detection techniques, while the proposed approach utilizes a machine learning algorithm. The algorithm consists of three key components: image preprocessing, image classification, and barcode interpretation. The researchers tested the performance of the machine learning-based algorithm and compared it to a conventional barcode scanner algorithm and commercial scanners. The results showed that the machine learning-based approach significantly outperformed the other methods, achieving higher accuracy rates, even when the barcode images were degraded with various types of noise and defects.
Applications
- Barcode scanning
Advantages
The incorporation of machine learning algorithms can lead to significant advancements in the field of barcode scanning, with the potential to improve processing times, increase scanner read rates, and reduce hardware requirements in industrial settings, such as delivery services. The improved read rates achieved by the machine learning-based approaches, even in the presence of image degradations, suggest that these techniques could be valuable tools for enhancing the performance of barcode scanners and driving innovation in this field.
Invention Readiness
This technology is at the protype level. The key components of the proposed approach include image preprocessing, image classification, and barcode interpretation. The machine learning-based approach achieved average accuracies of over 97% for the Number Set A and Number Set C classifiers, and an overall accuracy of over 99% when tested on a set of 10,000 generated UPC-A barcodes, significantly outperforming the 65% accuracy of the conventional barcode scanner algorithm. The machine learning-based approach also maintained high accuracy even when the barcodes were subjected to simulated defects and low contrast, showcasing its robustness.
IP Status
https://patents.google.com/patent/US11334731B2
