Deep Learning based Order Form Recognition

In this paper, a deep learning-based method for recognizing order forms is proposed. Firstly, a dataset is generated for classifying order forms. Using the YOLOv8 model, it is aimed to classify four different types of markings on the order forms. Thanks to proposed method high classification performance, up to 100%, under real-time inference conditions is achieved. The performance of different YOLOv8 model architectures for order forms are evaluated in terms of accuracy and inference speed. Additionally, our proposed preprocessing technique effectively extracts relevant areas from the forms, which enables the extraction of detailed order information.