基于改进YOLOv7⁃Tiny的高速公路入口两轮车辆闯入检测  

Two wheeled vehicle intrusion detection at highway entrance based on improved YOLOv7⁃Tiny

作　　者：王宏[1] 田恬 Wang Hong;Tian Tian(School of Computer Science,Xi’an Shiyou University,Xi’an 710065,China)

机构地区：[1]西安石油大学计算机学院,西安710065

出　　处：《现代计算机》2024年第8期17-23,共7页Modern Computer

摘　　要：近年来,浙江、福建等省区相继出台相关地方性法规,禁止两轮车辆(摩托车、电动车等)通行高速公路。针对高速公路入口工作人员难以实时检测到两轮车辆闯入的问题,提出一种改进YOLOv7⁃Tiny的两轮车辆闯入检测算法。首先,从VOC2005中提取摩托车图片并增补了带有入口背景的图片后形成新数据集;其次基于YOLOv7⁃tiny,引入ECA注意力机制,使模型更加聚焦训练摩托车相关目标特征。使用ssFPN网络,对小目标特征信息进行增强;采用基于动态非单调机制的WIoU损失函数,提高对于小物体检测的准确性;使用Adam优化器,提升回归过程的收敛速度和准确性。改进后的算法,mAP、Precision、Recall分别提高了2.63、4.01、13.92个百分点,F1提高0.10,表明该方法具有显著的有效性。In recent years,provinces such as Zhejiang and Fujian have successively introduced relevant local regulations pro⁃hibiting two wheeled vehicles(such as motorcycles,electric vehicles,etc.)from passing through highways.A modified YOLOv7⁃Tiny two wheeled vehicle intrusion detection algorithm is proposed to address the issue of real⁃time detection of two wheeled vehicle in⁃trusion by highway entrance workers.Firstly,motorcycle images were extracted from VOC2005 and images with entrance back⁃grounds were added to form a new dataset.Secondly,based on YOLOv7⁃Tiny,an ECA attention mechanism was introduced to make the model more focused on training motorcycle related target features.The ssFPN network was used to enhance small target feature information,and a WIoU loss function based on dynamic non monotonic mechanism was used to improve the accuracy of small ob⁃ject detection.Finally,use the Adam optimizer to improve the convergence speed and accuracy of the regression process.The im⁃proved algorithm improves mAP,Precision,Recall by 2.63,4.01 and 13.92 percentage point,respectively,and improves F1 by 0.10,indicating significant effectiveness of the method.

关 键 词：两轮车辆闯入检测 YOLOv7⁃tiny ECA注意力机制 ssFPN WIoU 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]