Automated Invoice Field Detection and Data Extraction Using YOLOv11 and PaddleOCR

Abstract

This study presents an integrated approach for automatically extracting and structuring information from invoices, captured as scanned documents or photographs, by leveraging a combination of object detection and optical character recognition (OCR). The primary objective was to develop a robust and efficient tool capable of accurately identifying and extracting relevant textual data from invoice images. This system minimizes manual data entry, reduces human error, and significantly enhances the speed and reliability of data processing workflows. The methodology involves training a custom object detection model to locate key invoice fields such as invoice number, date, GST details, and itemized amounts, followed by OCR techniques to extract and interpret the text within the detected regions. By combining these technologies, the solution demonstrates the ability to handle a wide range of invoice formats, layouts, and image qualities, ensuring adaptability and resilience across varied real-world scenarios. Despite the heterogeneous nature of the invoice templates used in the dataset, the proposed solution serves as a scalable and extensible framework for document information extraction. It can be seamlessly adapted to support additional document types beyond invoices, such as receipts, purchase orders, and contracts. Furthermore, the architecture is language-agnostic, with the potential to integrate multilingual OCR engines, making it applicable in diverse linguistic and geographic contexts. Overall, this study illustrates a practical and intelligent automation approach that can be employed in Enterprise Resource Planning (ERP), accounting software, and digital archiving systems to streamline document processing pipelines.