一种利用虚拟数据学习的电力部件识别方法

摘要

针对电力领域高质量标记数据不足的情况,提出了一种虚拟电力场景的搭建及相应的数据和标记的快速生成方法.并且本文通过虚拟数据集,以防振锤作为主要对象,经过3种基于机器学习的目标检测方法的实验比较与理论分析,得出可变部件模型(Deformable Parts Model,DPM)在虚拟数据集上训练后,能够在真实数据集取得了最好结果的结论.通过本文方法,可以快速得到针对电力线路部件的训练样本,并可以得到相对较好的初始分类结果,以进一步扩充样本以获得更精确的检测模型.同时本文的方法还可以方便的应用到其他电力场景与其他电力部件中去.%To cope with the problem of lack of labled image dataset in power line facility,this paper proposed a simple way to produce virtual image data and lable for power line scenario.This paper has experimented 3 different methods in machine learning object detection based on virtual dataset of damper,and the result shows that Deformable Parts Model (DPM) yield the best performance.With the method this paper presented,a training dataset and a detector can be easily obtained.Thus the primary result produced by the detector with a relatively good performance can help expand the training set and achieve a better detection model.Moreover,this method can be smoothly transferred to another power line facility and another power line scenario.