Exploring Automated Validation of Customs Documents with OCR and Language Models

This exploratory project, conducted in the fall of 2024, was ReLU’s first collaboration with NORBIT. The aim was to investigate how machine learning techniques could support the validation of customs declaration documents.

We explored various OCR tools and evaluated how large language models could be applied to interpret and structure raw textual data extracted from scanned forms. This included assessing the potential of fine-tuning models for improved document understanding and formatting. We also looked into techniques for classifying different regions of text based on layout and visual structure.

The project offered valuable insights into the feasibility of combining OCR and LLMs for document analysis.

  • NORBIT is a global provider of tailored technology solutions, structured across three core segments - Oceans, Connectivity, and Product Innovation & Realization (PIR). They support maritime markets with advanced sonar and sub-sea systems, offer smart wireless solutions for identification and tracking, and provide R&D, prototyping, and manufacturing services. Headquartered in Trondheim, Norway, with additional facilities in the US, Hungary, Selbu and Røros, NORBIT employs around 450–600 people globally and maintains a worldwide sales and distribution platform.