This paper presents a deep learning-based end-to-end system for the automatic extraction of key information from structured certificate documents. The model was trained with 1,784 manually labeled certified corpus images. The research finds locations in the dataset utilizing the YOLO models v11 and v12. While YOLOv11 has precision = 0.987, recall = 0.996, mAP@50 = 0.981, and mAP@50–95 = 0.678, YOLOv12 has precision = 0.992, recall = 0.998, mAP@50 = 0.986, and mAP@50–95 = 0.690. It can be seen from the experimental results that YOLO v12 excels in detecting objects (19.1 ms vs. 224.4 ms per image). In order to realize the capability to extract and verify good certificate information, the research then proposes an integrated object detection, optical character recognition (OCR), and database comparison process. In the future, instance segmentation, multimodal learning, and personalized OCR enhancement can be employed to further improve the system's performance on different categories of documents.

Information Extraction Object Detection Deep Learning OCR