[1]石露露,廖光忠.改进 YOLOv5s 的明渠漂浮垃圾实时检测方法[J].计算机技术与发展,2023,33(09):83-90.[doi:10. 3969 / j. issn. 1673-629X. 2023. 09. 013]
　SHI Lu-lu,LIAO Guang-zhong.Real-time Detection Method of Floating Garbage in Open Channels Based on Improved YOLOv5s[J].,2023,33(09):83-90.[doi:10. 3969 / j. issn. 1673-629X. 2023. 09. 013]

点击复制

改进 YOLOv5s 的明渠漂浮垃圾实时检测方法()

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

33

期数:

2023年09期

页码:

83-90

栏目:

人工智能

出版日期:

2023-09-10

文章信息/Info

Title:

Real-time Detection Method of Floating Garbage in Open Channels Based on Improved YOLOv5s

文章编号:

1673-629X(2023)09-0083-08

作者:

石露露¹ ; 廖光忠²

1. 武汉科技大学 计算机科学与技术学院,湖北 武汉 430065;

2. 武汉科技大学 智能信息处理与实时工业系统湖北省重点实验室,湖北 武汉 430065

Author(s):

SHI Lu-lu1 ; LIAO Guang-zhong2

1. School of Computer Science and Technology,Wuhan University of Science and Technology,Wuhan 430065,China;

2. Hubei Provincial Key Laboratory of Intelligent Information Processing and Real-time Industrial System,Wuhan University of Science and Technology,Wuhan 430065,China

关键词:

YOLOv5s; 实时检测; 加权双向特征融合; 注意力机制; 小目标

Keywords:

YOLOv5s; real-time detection; weighted bidirectional feature fusion; attention mechanism; small objects

分类号:

TP391

DOI:

10. 3969 / j. issn. 1673-629X. 2023. 09. 013

摘要:

针对航拍图像上明渠漂浮垃圾尺寸小,且易受水面倒影、强光等因素干扰从而造成漂浮垃圾漏检和误检的问题,提出了一种基于改进 YOLOv5s 的明渠漂浮垃圾实时检测

方法。 首先,通过数据增强的方式对数据集进行扩充,避免数据量过少产生过拟合现象;然后,结合加权双向特征融合网络( BiFPN) ,对 YOLOv5s 结构的特征融合过程进行

修改,以提高检测精度和速度;并且在 Neck 和 Head 部分之间添加 3 个改进的三维 CBAM 注意力机制模块,加强网络信息的提取和定位能力,能够有效地降低检测的漏检率和误检率;最后,增大网络输入的分辨率,使图像具有更加丰富的细节信息和更加精确的位置信息,有利于小目标特征信息的提取。 实验结果显示改进的 YOLOv5s 算法检测的平均精度达到了 89. 9% ,与原 YOLOv5s 算法相比提高了 9. 5% ,而且与其他目标检测算法对比,能够提高明渠漂浮垃圾检测的精度,确保检测的实时性。

Abstract:

Aiming at the problem that floating garbage in open channels is small in size and easily disturbed by factors such as reflectionon the water surface and strong light,　resulting in missed detection and false detection of floating garbage,a real-time detection method offloating garbage in open channels based on improved YOLOv5s is proposed. Firstly, the data set is expanded by means of dataenhancement to avoid overfitting caused by too little data. Then, combined with the weighted bidirectional feature pyramid network( BiFPN) ,the feature fusion process of the YOLOv5s structure is modified to improve the detection accuracy and speed. Next, threeimproved 3D CBAM attention mechanism modules are added between the Neck and Head parts to enhance the extraction and positioningcapabilities of network information,　which can effectively reduce the missed detection rate and false detection rate of detection. Finally,the network input resolution is increased,so that the image had richer detailed information and more accurate position information,whichis conducive to the extraction of small target feature information. The experimental results show that the average detection accuracy of theimproved YOLOv5s algorithm reaches 89. 9% ,which is 9. 5% higher than that of the original YOLOv5s algorithm,and compared withother target detection algorithms,it can improve the detection accuracy and ensure the real-time detection.

常用功能