合肥工业大学校徽

合肥工业大学学报自科版

合肥工业大学学报(自然科学版)

JOURNAL OF HEFEI UNIVERSITY OF TECHNOLOGY (NATURAL SCIENCE)

主管：中华人民共和国教育部

主办：合肥工业大学

ISSN 1003-5060, CN 34-1083/N

基于改进 YOLOv5 算法的票据检测

Bill detection based on improved YOLOv5 algorithm

期刊信息

合肥工业大学（自然科学版），2024年11月，第47卷第11期：1459-1464

DOI: 10.3969/j.issn.1003-5060.2024.11.004

作者信息

扈静，贺竞娇，龚宇，汪俊峰

（合肥工业大学机械工程学院，安徽合肥230009）

摘要和关键词

摘要: 在票据全流程识别过程中, 针对不同类型票据的目标检测是关键步骤, 为实现日常报销过程中不同种类票据快速检测其类型和位置, 文章提出一种基于改进 YOLOv5 算法的票据检测方法。对于原始数据集进行预处理, 模拟票据检测中可能出现的干扰信息, 提高训练模型的性能; 利用 CSPDarkNet53 网络进行特征提取, 采用基于重合面积、中心点距离、长宽比和角度 4 个几何参数的 SIoU 目标位置损失函数, 对原损失函数进行改进, 提升 YOLOv5 票据检测的精度, 构建票据检测方法。最后通过自建实际拍摄的票据数据集来验证基于改进 YOLOv5 算法的票据检测的有效性。结果表明, 该方法拥有较高的平均精度均值 (99.20%) 和检测速度 (51 帧/s), 可以满足实际应用场景的要求。

关键词: 票据；目标检测；识别；改进YOLOv5；损失函数

Authors

HU Jing, HE Jingjiao, GONG Yu, WANG Junfeng

(School of Mechanical Engineering, Hefei University of Technology, Hefei 230009, China)

Abstract and Keywords

Abstract: In the whole process of bill recognition, object detection for different types of bills is a key step. In order to realize the rapid detection of the types and positions of different types of bills in the daily reimbursement process, this paper proposes a bill detection method based on the improved YOLOv5 algorithm. The collected images are preprocessed to simulate the possible interference information in the bill detection and improve the robustness of the training model. With the help of CSPDarkNet53 network for feature extraction, the SIoU target position loss function based on the four geometric parameters of coincidence area, center point distance, aspect ratio and angle is used to improve the original loss function, so as to improve the accuracy of YOLOv5 bill detection, and build a bill detection model. Finally, the effectiveness of the bill detection based on the improved YOLOv5 algorithm is verified through the actual test set. The experimental results show that the method has high mean average precision (99.20%) and detection speed (51 frames per second), and can meet the requirements of practical application scenarios.

Keywords: bill; object detection; recognition; improved YOLOv5; loss function

基金信息

安徽省科技攻关计划资助项目（JZ2016AKKG0837）

前一篇后一篇

返回期刊目录