Document Layout Classification and Analysis Using YOLOv8

Document layout classification is crucial for digitizing documents, enabling artificial intelligent (AI) models to understand and process information from diverse sources. (1) This study investigates the application of You Only Look Once version 8 (YOLOv8), a state-of-the-art object detection model, for efficient document layout classification. (2) The objective is to evaluate YOLOv8's feasibility in accurately identifying key document components. The author converts labeled data to the YOLOv8 format, trains the model with optimized settings on Apple Silicon, and validates it. YOLOv8 demonstrates promising performance on the validation set, achieving a mean Average Precision (mAP)@0.5 of 0.916 (3) High precision in detecting classes are like text (0.88) and title (0.93). The results confirm that YOLOv8 offers fast inference speeds and handles limited datasets robustly, but needs to improve its data augmentation strategy for rare classes. (4) This approach facilitates automated document digitization and enhances machine understanding of document structures. Future research will focus on data enhancement strategies to extend the dataset and drive more efficient intelligent document processing.