首页 | 本学科首页   官方微博 | 高级检索  
     

改进的YOLOv3安检包裹中危险品检测算法
引用本文:卢官有,顾正弘. 改进的YOLOv3安检包裹中危险品检测算法[J]. 计算机应用与软件, 2021, 38(1): 197-204. DOI: 10.3969/j.issn.1000-386x.2021.01.033
作者姓名:卢官有  顾正弘
作者单位:扬州大学信息工程学院 江苏 扬州 225127;扬州大学信息工程学院 江苏 扬州 225127
摘    要:针对目前的安检工作中检测效率较低的问题,提出一种改进的YOLOv3安检包裹中危险品检测算法,用于辅助安检人员完成安检工作,提升安检机的自动化和智能化。将YOLOv3中原来每个网格预测3个边界框减少到预测2个边界框,检测速度提升了约6%。利用K-means聚类根据数据集计算出先验框,平均精度均值提高了约1.13个百分点。为了解决样本量少的问题,采用数据增强方法,平均精度均值提升了约7.8个百分点。采用多尺度输入训练策略,不仅增强了模型检测不同尺度图像时的鲁棒性,而且平均精度均值提升了约1.22个百分点。

关 键 词:YOLOv3  数据增强  危险品检测  K-MEANS  卷积神经网络

A DANGEROUS GOODS DETECTION ALGORITHM BASED ON IMPROVED YOLOv3
Lu Guanyou,Gu Zhenghong. A DANGEROUS GOODS DETECTION ALGORITHM BASED ON IMPROVED YOLOv3[J]. Computer Applications and Software, 2021, 38(1): 197-204. DOI: 10.3969/j.issn.1000-386x.2021.01.033
Authors:Lu Guanyou  Gu Zhenghong
Affiliation:(College of Information Engineering,Yangzhou University,Yangzhou 225127,Jiangsu,China)
Abstract:Aiming at the low efficiency of detection in current security inspections,this paper proposes an improved target detection algorithm based on YOLOv3 for security inspection problems.It is helpful for auxiliary security staff to complete the security inspection work,and it improves the automation and intelligence of the security inspection machine.It changed the original predict three bounding box of each grid cell in YOLOv3 to two bounding boxes,and the detection speed was improved by approximately 6%.Based on our own datasets,the K-means clustering was used to calculate the anchors according the data base,and the mAP was increased by 1.13 percentage point.In order to solve the problem of small sample size of datasets,this paper used data augmentation method,and mAP was increased by 7.84 percentage point.The multi-scale input training strategy was adopted in this paper,which enhanced the robustness of the model when detecting images of different scales,and the mAP was increased by 1.12 percentage point.
Keywords:YOLOv3  Data augmentation  Dangerous goods detection  K-means  Convolutional neural network
本文献已被 维普 万方数据 等数据库收录!
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号