摘要
针对输电线路小金具缺失的检测问题,对小金具缺失算法的加速推理进行研究,采用多任务头的学习方法,将小金具缺失检测任务使用一个Swin Transformer网络结构[12]和多个由多层感知机组成的任务头组合的方式进行多任务学习推理,并进行单任务和多任务学习的推理精度和推理性能对比实验,最后还通过实验得到在多任务学习中插拔式扩展任务,实验结果表明在输电线路小金具缺失推理算法中多任务学习比单任务学习的推理性能提升了2倍多,同时显存占用降低了22%以上;通过插拔式扩展任务实验,验证了扩展任务的效果,可灵活扩展配置任务。
Aiming at the detection problem of missing small fittings in transmission lines,the acceleration algorithm of the missing small fittings is studied by the multi-task learning.The missing small fittings detection task is connected by a Swin Transformer backbone network and multiple MLP task heads,a new approach with the multi-task learning and multi-task inference is carried out,and compared with the accuracy and performance of the single-task learning and multi-task learning.It is verified that the extended tasks can be seamlessly increased in the multi-task learning framework.The experimental results show that the speed of the proposed multi-task learning is over 2 times higher than that of the single-task learning,and the memory usage is reduced by over 22%.The experimental results of the extended tasks validate the effectiveness of the proposed multi-task learning framework and the flexibility of the task settings.
作者
程绳
葛雄
肖非
朱传刚
吴军
肖海涛
李嗣
楚江平
袁雨薇
CHENG Sheng;GE Xiong;XIAO Fei;ZHU Chuangang;WU Jun;XIAO Haitao;LI Si;CHU Jiangping;YUAN Yuwei(State Grid Hubei Extra High Voltage Company,WuHan 430050,China;Binjiang Institute of Zhejiang University,HangZhou 31000,China)
出处
《计算机测量与控制》
2023年第7期251-257,共7页
Computer Measurement &Control
关键词
多任务头学习
加速推理
输电线路
小金具缺失
扩展任务学习
multi-task learning
inference acceleration
transmission line
missing small fittings
extended task learning
