University of Twente Student Theses
Product Recognition in Store Environments : A Deep Learning Approach
Janssen, D. (2024) Product Recognition in Store Environments : A Deep Learning Approach.
This is the latest version of this item.
|
PDF
15MB |
| Abstract: | Product recognition technology is a promising strategy to revolutionize retail operations, with significant implications for shelf and inventory management, automated checkout, and loss prevention. Despite its potential, accurately identifying products in images remains challenging due to the complexities of retail environments. This study investigates the effectiveness of deep learning for product recognition in store environments. To this end, we designed and validated a new deep learning pipeline which builds upon existing works in the field and is driven by practical considerations. Our method involves detecting products at the contour level through instance segmentation, followed by classification of the detected products using an embedding model trained with novel example mining strategies. The generalizability of our approach was assessed through cross-dataset evaluation, and an evaluation with stakeholders was performed to assess its potential for practice. Our results indicate that the use of example mining strategies to train the classification model with informative samples significantly improved the accuracy (K=1 from 93.1% to 96.3% and 80.4% to 85.8% for the internal and Grocery products dataset, respectively). In addition, the suppression of background regions in the image from contour-based localization also enhanced the classification performance for the internal dataset (K=1 from 94.1% to 96.3%). Overall, the full product recognition pipeline performs well on images similar to the training data (mAP = 85.0%), and shows potential for generalizing to different product assortments and store environments (mAP = 60.8%). While further work is needed to improve robustness against variations in image conditions such as scale and position, our deep learning approach shows promising results for product recognition, a conclusion supported by positive feedback from stakeholders. |
| Item Type: | Essay (Master) |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Subject: | 54 computer science, 85 business administration, organizational science |
| Programme: | Business Information Technology MSc (60025) |
| Link to this item: | https://purl.utwente.nl/essays/103711 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page