基于样本扩充的Faster R-CNN电网异物监测技术

Research on Foreign Matter Monitoring of Power Grid With Faster R-CNN Based on Sample Expansion

电网公司的巡检工作主要依靠人工完成,需要大量人力物力,且实时性较差。针对该问题,提出一个基于区域推荐卷积神经网络的图像目标监测系统,其中核心算法为Faster R-CNN算法。利用深度学习对摄像装置所采集的现场图像进行分析,执行目标检测任务,若发现威胁电网安全运行的隐患将及时通知工作人员。深度学习发挥其优势需要有效样本达到一定数量,包含隐患的真实样本较少,有些异物种类甚至没有合适的样本,往往不能满足深度学习算法的训练要求。因此研究了一种用于扩充样本的样本生成算法,将隐患目标与背景图像按照一定规则进行融合,达到批量扩充样本集的目的。使用该算法生成的样本进行测试,测试结果表明扩充后的训练集可以使系统性能得到一定提升。此外,通过测试发现,对训练集做一定的预处理可以提升模型的识别性能。

Some substations and transmission lines are close to residential areas or construction sites,which may threaten safe operation of power grid.At present,most of on-site inspections are carried out manually,spending a lot of manpower and time,and might be poor in real-time performance.In order to solve this problem,an image target monitoring system based on region proposal convolutional neural network is proposed.This system based on deep learning is used to analyze the scene images collected by camera equipment.It will timely inform the persons in charge when hidden dangers are found.Deep learning requires a number of effective samples to work.But there are few real samples of hidden dangers in actual operation,which cannot meet the demand.Therefore,a sample generation algorithm to expand the sample is proposed.And the digital image processing technology is used to preprocess and further expand the training set.Experiment results show that the performance of the proposed model is improved obviously by using the extended training set.