Deep Learning Framework for Detecting, Classifying, and Recognizing Invoice Metadata

Summary

In general, images or invoice images, are rich information sources; however, processing large amounts of invoices in finance requires extensive human work and material resources. The demand for a technological tool to process invoices is essential. Therefore, this study proposes a framework based on deep learning techniques for recognizing specific invoice information (or metadata). The deep learning model used in this research includes the classic architecture of Convolutional Neural Networks and its other variations, such as VGG‐16 and Faster R‐CNN. The complete process includes image processing series, model training for field detection and classification, and optical character recognition (OCR) for the detected fields. The proposed framework collects relevant information from invoice images and converts them into a text format to facilitate invoice management and transform them into structured data. The framework performance shows a significant value in recognizing invoice metadata compared with YOLO v4.