军事知识服务平台
基于改进DeepSORT的接触网绝缘子多目标跟踪冲洗算法
Multi-Target Tracking Flushing Algorithm for Catenary Insulator Based on Improved DeepSORT
摘要 针对现有绝缘子水冲洗机器人在接触网多目标环境下进行移动冲洗作业时存在的目标丢失、跟踪延迟以及冲洗紊乱等问题,提出了一种基于改进DeepSORT的接触网绝缘子多目标跟踪冲洗算法.首先,检测端采用改进后的YOLOv5s模型;其次,针对跟踪过程中参数过多、计算量较大的问题,设计了一种新的轻量化网络架构MSNet(Mobile SimAM Network),实现对特征提取模型的大幅压缩;最后,优化DeepSORT目标ID赋值机制,并搭建机器人验证.试验结果表明:改进算法相较于原始DeepSORT算法,其HOTA、MOTP、IDF1 分别提升了 16.669%、6.924%、14.230%,ID切换次数仅为1 次,在低算力计算机上运行速率可达39 f/s.本研究算法可满足绝缘子冲洗作业的准确度及实时性要求,并具有一定的实际应用性.
Abstract:
In order to solve the problems such as target loss,tracking delay and flushing disorder of the existing insulator water flushing robot in the multi-target environment of catenary,an improved Deep-SORT multi-target tracking flushing algorithm for catenary insulators was proposed.Firstly,the im-proved YOLOv5s model was adopted in the detection terminal.Secondly,a new lightweight Network ar-chitecture,MSNet(Mobile SimAM Network),is designed to compress the feature extraction model.Fi-nally,the DeepSORT target ID assignment mechanism is optimized,and robot verification is built.The experimental results show that compared with the original DeepSORT algorithm,the improved algorithm improves HOTA,MOTP and IDF1 by 16.669%,6.924%and 14.230%respectively.The number of ID switching is only one time,and the running speed on the low-power computer can reach 39 f/s.The al-gorithm proposed can meet the requirements of accuracy and real-time in insulator flushing operation,and has certain practical applicability.
刊 名 电瓷避雷器
年,卷(期): 2026, (5)
机标分类号: TP242 TP393 TN929.5
在线出版日期: 2026年11月9日