首页|基于改进YOLOv5的港口集装箱损伤检测算法

基于改进YOLOv5的港口集装箱损伤检测算法

Damage detection algorithm of port container based on improved YOLOv5

扫码查看
原文链接
  • NETL
  • NSTL
  • 万方数据
针对港口复杂环境背景下,不同尺度间多种类集装箱损伤目标检测精度低的问题,提出一种基于改进YOLOv5的港口集装箱损伤检测算法.通过使用一维卷积改进卷积块注意力机制(CBAM)中空间注意力模块的池化操作,后设计残差结构,并验证在不同位置引入改进的CBAM基本块对模型性能的影响,探索尽量减小复杂背景对检测结果影响的最佳方案及融合位置;为有效解决不同集装箱损伤图像尺度特征变换较大的问题,依据双向特征融合网络(Bi-FPN)结构思想,对颈部特征融合网络进行改进,在不过多增加计算量的情况下,更好地增强网络对多尺度目标的特征融合能力;最后将EIOU Loss替换GIOU Loss作为算法的损失函数,在降低算法边界框回归损失的同时提高算法的检测精度.实验结果表明,改进YOLOv5算法的平均检测精度达到了98.32%,较原YOLOv5目标检测算法提高了4.28%,同时保证了检测速度,验证了所提出算法的有效性,对港口企业高精度验箱的工业部署有重要意义.
Aiming at the problem of low detection accuracy of multiple types of container damage targets at different scales in the complex port environment,a port container damage detection algorithm based on improved YOLOv5 is proposed.By using one-dimensional convolution to improve the pooling operation of spatial attention module in CBAM attention mechanism,the residual structure was designed,and the influence of introducing improved CBAM blocks in different positions on the model performance was verified,to explore the best scheme and fusion location to minimize the influence of complex background on detection results.In order to solve the problem of significant change in feature transformation of different container damage images,the neck feature fusion network was improved based on the Bi-FPN network structure,it can enhance the feature fusion ability of the network to multi-scale targets without increasing the amount of calculation too much.Finally,EIOU Loss was used to replace GIOU Loss as the loss function of the algorithm,which reduced the regression loss of the bounding box and improved the detection accuracy of the algorithm.The experimental results show that the average detection accuracy of the improved YOLOv5 model have reached 98.32%,which is 4.28%higher than the original YOLOv5 algorithm.At the same time,it ensures the detection speed and verifies the validity of the algorithm proposed in this paper,which is of great significance to the industrial deployment of high-precision container inspection in port enterprises.

container damagetarget detectionYOLOv5CBAMBi-FPNEIOU

裴晓芳、刘菁宇、柏雪、周进、衡晓钰

展开 >

无锡学院电子信息工程学院 无锡 214105

南京信息工程大学电子与信息工程学院 南京 210044

集装箱损伤 目标检测 YOLOv5 CBAM Bi-FPN EIOU

国家自然科学基金委青年基金高校哲学社会科学研究一般项目苏高教会"高质量公共课教学改革研究"专项课题无锡学院教改课题教育部产学合作协同育人项目教育部产学合作协同育人项目2022江苏省大学生创新创业训练计划

422050782022SJYB09792022JDKT138JGZD202107-3-11220703806203838220702116135122202213982054Y

2024

10.19652/j.cnki.femt.2305518

国外电子测量技术
北京方略信息科技有限公司

国外电子测量技术

CSTPCD
影响因子：1.414
ISSN：1002-8978
年,卷(期)：2024.43(2)
  • 27