基于YOLOv8n的甲骨文图像识别技术研究

Research on Oracle Bone Script Image Recognition Technology Based on YOLOv8n

摘要: 随着深度学习技术的发展，物体检测算法在图像识别领域取得了显著进展。针对甲骨文图像识别技术的研究，本文基于YOLOv8n框架构建了一种高效的甲骨文物体检测与分类方法。通过数据预处理、数据增强、模型训练、模型预测等环节，结合CSPDarknet骨干网络、PAFPN特征金字塔和CIoU损失函数，提升了检测精度。该验证结果表明目标检测模型的P值和召回率R分别达到0.855和0.807，其中mAP@0.5的指标达到0.876，mAP@[0.5:0.95]的值为0.53。分类模型的准确率高达93.6%，比YOLOv5提高了6.6%。本研究为甲骨文的数字化与智能分析提供了新的技术方案，探索了提高识别与分类精度及泛化能力的方法。

Abstract: With the development of deep learning technology, object detection algorithms have achieved significant progress in the field of image recognition. Focusing on the recognition of oracle bone script images, this paper proposes an efficient object detection and classification method based on the YOLOv8n framework. Through stages such as data preprocessing, data augmentation, model training, and prediction, and by integrating the CSPDarknet backbone, PAFPN feature pyramid, and CIoU loss function, the detection accuracy is effectively improved. Experimental results show that the object detection model achieves a precision (P) of 0.855 and a recall (R) of 0.807, with an mAP@0.5 of 0.876 and an mAP@[0.5:0.95] of 0.53. The classification model attains an accuracy of 93.6%, representing a 6.6% improvement over YOLOv5. This study provides a novel technical solution for the digitization and intelligent analysis of oracle bone inscriptions, and explores methods to enhance recognition accuracy, classification performance, and model generalization capability.