摘要
针对现有绝缘子缺失检测模型计算复杂度高和小目标难以检测等问题,提出一种基于改进的YOLOv5s轻量级检测模型。首先,移除主干网络中的C3模块来减少模型的参数量。其次,在多尺度特征融合网络中引入卷积块注意力机制来提高复杂背景下模型的特征提取能力。同时,采用加权双向特征金字塔网络结构对特征进行双向跨尺度加权融合,提升网络在遮挡物、相似目标干扰下目标的检测性能。最后,选用SIoU损失函数提升网络的收敛速度和检测精度。实验结果表明,所提模型的平均精准率为96.8%,浮点运算数为2.8 GFLOPS,而原始YOLOv5s在保证97.4%的平均精准率下的浮点运算数为16.3 GFLOPS。相较于原始模型,所提模型对小目标、遮挡目标以及模糊等场景有着较强的鲁棒性,且在保证近似检测精度的同时极大减少了计算量。
Aiming at the problems of the existing insulator missing detection model,such as large parameters and difficult detection of small targets,an improved lightweight detection model based on YOLOv5s is proposed.Firstly,the C3 module in the backbone network is removed,which reduces the number of parameters of the model.Secondly,the convolutional block attention module is introduced into the multiscale feature fusion network,which improves the feature extraction ability of the network in complex backgrounds.Then,the weighted bi-directional feature pyramid network structure is used to perform a bidirectional cross-scale weighted fusion of features to improve the detection performance of targets under the interference of obstructions and similar targets.Finally,the SIoU loss function is selected to improve the convergence speed and detection accuracy.The experimental results show that the average accuracy of the proposed model is 96.8%,and the floating-point number of operations is 2.8 GFLOPS,while the original YOLOv5s floating-point number of operations is 16.3 GFLOPS with an average accuracy of 97.4%.Compared with the original model,the proposed model is more robust to scenarios such as small targets,interference of obstructions and similar targets,and greatly reduces the amount of computation while ensuring approximate detection accuracy.
作者
池小波
张伟杰
贾新春
续泽晋
CHI Xiaobo;ZHANG Weijie;JIA Xinchun;XU Zejin(School of Automation and Software,Shanxi University,Taiyuan 030013,China;School of Mathematical Sciences,Shanxi University,Taiyuan 030006,China)
出处
《测试技术学报》
2024年第1期19-26,共8页
Journal of Test and Measurement Technology
基金
国家自然科学基金资助项目(61803244,61973201)
山西省基础研究计划资助项目(202203021211297)
山西省回国留学人员科研资助项目(2023-002)。
