Högskolan i Skövde

his.sePublications
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • apa-cv
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
NLP Cross-Domain Recognition of Retail Products
University of Skövde, School of Engineering Science. University of Skövde, Virtual Engineering Research Environment. Department of Computing, Jönköping AI Lab, Jönköping University, Sweden ; ITAB Shop Products AB, Jönköping. (Virtual Production Development (VPD))ORCID iD: 0000-0001-8880-7965
Department of Computing, Jönköping AI Lab, Jönköping University, Sweden.
Department of Computing, Jönköping AI Lab, Jönköping University, Sweden.
2022 (English)In: ICMLT '22: Proceedings of the 2022 7th International Conference on Machine Learning Technologies, March 2022, Association for Computing Machinery (ACM), 2022, p. 237-243Conference paper, Published paper (Refereed)
Abstract [en]

Self‐checkout systems aim to provide a seamless and high-quality shopping experience and increase the profitability of stores. These advantages come with some challenges such as shrinkage loss. To overcome these challenges, automatic recognition of the purchased products is a potential solution. In this context, one of the big issues that emerge is the data shifting, which is caused by the difference between the environment in which the recognition model is trained and the environment in which the model is deployed. In this paper, we use transfer learning to handle the shift caused by the change of camera and lens or their position as well as critical factors, mainly lighting, reflection, and occlusion. We motivate the use of Natural Language Processing (NLP) techniques on textual data extracted from images instead of using image recognition to study the efficiency of transfer learning techniques. The results show that cross-domain NLP retail recognition using the BERT language model only results in a small reduction in performance between the source and target domain. Furthermore, a small number of additional training samples from the target domain improves the model to perform comparable as a model trained on the source domain.

Place, publisher, year, edition, pages
Association for Computing Machinery (ACM), 2022. p. 237-243
Keywords [en]
Image recognition, Learning systems, Natural language processing systems, Sales, Text processing, Transfer learning, BERT, Cross-domain, Domain adaptation, Language processing, Natural language processing, Natural languages, Product recognition, Retail, Text classification, Classification (of information), NLP
National Category
Computer Sciences
Research subject
Virtual Production Development (VPD)
Identifiers
URN: urn:nbn:se:his:diva-22974DOI: 10.1145/3529399.3529436ISI: 001053939400037Scopus ID: 2-s2.0-85132414844ISBN: 978-1-4503-9574-8 (print)OAI: oai:DiVA.org:his-22974DiVA, id: diva2:1780233
Conference
ICMLT 2022: 7th International Conference on Machine Learning Technologies (ICMLT), Virtual Conference, 11-13 March 2022
Funder
Knowledge Foundation, DATAKIND 20190194
Note

This work was supported by the Swedish Knowledge Foundation (DATAKIND 20190194), the company ITAB, and Smart Industry Sweden (KKS-2020-0044).

© 2022 Association for Computing Machinery.

Available from: 2023-07-05 Created: 2023-07-05 Last updated: 2025-12-15Bibliographically approved
In thesis
1. Product Recognition with OCR Text: Advancing Grocery Product Recognition through Robust Approaches, Fine-Grained Recognition, and Domain Adaptation for Real-Time Performance
Open this publication in new window or tab >>Product Recognition with OCR Text: Advancing Grocery Product Recognition through Robust Approaches, Fine-Grained Recognition, and Domain Adaptation for Real-Time Performance
2025 (English)Doctoral thesis, comprehensive summary (Other academic)
Abstract [en]

The physical retail sector faces challenges in improving operational efficiency, reducing costs, and enhancing customer experience. Over the past decade, companies have introduced product recognition technology solutions to improve checkout efficiency, inventory management, and fraud detection. However, most initiatives have struggled to scale or achieve sufficient accuracy due to the complex nature of physical retail, which includes a large number of products that continuously change, as well as varied environmental conditions. In parallel, academic research has tackled many of these challenges by providing datasets and new methods to improve recognition performance, but considerable challenges persist.

This thesis addresses three main challenges in grocery product recognition: robust recognition, recognition of visually similar products, and domain adaptation between different retail systems. To address these challenges, the work centers on the use of Optical Character Recognition (OCR) to extract textual information found on product packaging for product recognition. With extensive experiments and the creation of a dataset, the results show that OCR-based methods for product recognition can improve recognition robustness, enable more accurate differentiation between similar products, and also work across different retail systems.

Therefore, the main contribution of this thesis is the development and validation of these OCR text-based methods and approaches, specifically designed to address the requirements in physical retail.

Abstract [sv]

Den fysiska detaljhandeln står inför utmaningar med att förbättra den operativa effektiviteten, sänka kostnader och stärka kundupplevelsen. Under det senaste decenniet har företag introducerat lösningar baserade på produktigenkänningsteknik för att effektivisera kassaprocesser, förbättra lagerhantering och upptäcka bedrägerier. De flesta initiativ har dock haft svårt att skala upp eller uppnå tillräcklig noggrannhet på grund av detaljhandelns dynamiska natur, som kännetecknas av ett stort och ständigt föränderligt produktsortiment samt skiftande butiksmiljöer. Parallellt har akademisk forskning adresserat många av dessa utmaningar genom att tillhandahålla datamängder och nya metoder för att förbättra igenkänningsprestandan, men betydande utmaningar kvarstår.

Denna avhandling adresserar tre huvudsakliga utmaningar inom produktigenkänning i dagligvaruhandeln: robust igenkänning, finmaskig igenkänning av visuellt liknande produkter samt domänanpassning mellan olika retailsystem. För att hantera dessa utmaningar fokuserar arbetet på att använda optisk teckenläsning (OCR) för att extrahera textinformation från produktförpackningar för produktigenkänning. Genom omfattande experiment och skapandet av en datamängd visar resultaten att OCR-baserade metoder kan förbättra robustheten i igenkänningen, möjliggöra mer noggrann differentiering mellan produkter samt fungera över olika retailmiljöer.

Avhandlingens huvudsakliga bidrag är utvecklingen och valideringen av metoder och tillvägagångssätt för produktigenkänning med text från OCR som möter de unika kraven inom den fysiska detaljhandeln.

Place, publisher, year, edition, pages
Skövde: University of Skövde, 2025. p. xv, 140
Series
Dissertation Series ; 67
National Category
Computer graphics and computer vision Natural Language Processing Artificial Intelligence
Research subject
Virtual Production Development (VPD)
Identifiers
urn:nbn:se:his:diva-26062 (URN)978-91-989080-7-7 (ISBN)978-91-989080-8-4 (ISBN)
Public defence
2026-01-23, ASSAR Industrial Innovation Arena, Kavelbrovägen 2B, 541 36, Skövde, 09:15 (English)
Opponent
Supervisors
Note

Ett av sex delarbeten (övriga se rubriken Delarbeten/List of papers):

6. Tobias Pettersson, Maria Riveiro, and Tuwe Löfström. “Real-Time OCR-Based Grocery Product Recognition with Orientation Alignment and Embedding-Driven Classification”. In: Accepted and presented at the International Conference on Machine Vision (ICMV 2025). 2025.

PUBLICATIONS WITH LOW RELEVANCE

7. Puneet Mishra, Aneesh Chauhan, and Tobias Pettersson. “Seeing through plastics: A novel combination of NIR hyperspectral imaging and spectral orthogonalization for detecting fresh fruit inside plastic packaging to support automated barcode less checkouts in supermarkets”. In: Food Control 150 (2023), p. 109762.

8. Faeze Zakaryapour Sayyad, Tobias Pettersson, Seyed Jalaleddin Mousavirad, Irida Shallari, and Mattias O’Nils. “AdAPT: Advertisement detector adaptation under newspaper domain shift with null-based pseudo-labeling”. In: Machine Learning with Applications (2025), p. 100806.DOI: https://doi.org/10.1016/j.mlwa.2025.100806.

Available from: 2025-12-15 Created: 2025-12-12 Last updated: 2025-12-15Bibliographically approved

Open Access in DiVA

No full text in DiVA

Other links

Publisher's full textScopus

Authority records

Pettersson, Tobias

Search in DiVA

By author/editor
Pettersson, Tobias
By organisation
School of Engineering ScienceVirtual Engineering Research Environment
Computer Sciences

Search outside of DiVA

GoogleGoogle Scholar

doi
isbn
urn-nbn

Altmetric score

doi
isbn
urn-nbn
Total: 360 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • apa-cv
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf