| 引用本文: | 龚钢军,张帅,吴秋新,陈志敏,刘韧,苏畅.基于TensorFlow的高压输电线路异物识别[J].电力自动化设备,2019,39(4): |
| GONG Gangjun,ZHANG Shuai,WU Qiuxin,CHEN Zhimin,LIU Ren,SU Chang.Foreign body identification based on TensorFlow for high voltage transmission line[J].Electric Power Automation Equipment,2019,39(4): |
|
| 摘要: |
| 针对传统异物识别准确率较低的问题，提出一种基于TensorFlow的深度卷积神经网络的异物识别模型。将巡检图像进行图像灰度化和尺寸压缩等预处理，并采用三维块匹配滤波(BM3D)算法进行图像去噪得到实验所需的训练数据。提出基于TensorFlow的深度卷积神经网络框架，通过使用框架中的TensorBoard模块设计深度卷积神经网络模型结构与优选模型参数，并针对ReLU激活函数与特征权重进行理论分析。实验结果表明，经过15次迭代训练后，深度卷积神经网络比传统的支持向量机(SVM)、极限学习机(ELM)和BP神经网络算法具有更强的巡检图像识别能力；与经典的LeNet-5和VGGNet模型以及相关文献中的模型相比，所提模型更具有优越性。 |
| 关键词: 输电线路 异物识别 卷积神经网络 TensorFlow |
| DOI：10.16081/j.issn.1006-6047.2019.04.030 |
| 分类号:TM761 |
| 基金项目: |
|
| Foreign body identification based on TensorFlow for high voltage transmission line |
|
GONG Gangjun1, ZHANG Shuai2, WU Qiuxin2, CHEN Zhimin1, LIU Ren3, SU Chang1
|
|
1.Beijing Engineering Research Center of Energy Electric Power Information Security, North China Electric Power University, Beijing 102206, China;2.School of Applied Science, Beijing Information Science & Technology University, Beijing 100192, China;3.Beijing Excellent Network Security Technology Corp.,Ltd.,Beijing 102206, China
|
| Abstract: |
| Aiming at the low accuracy rate of traditional foreign body identification, a foreign body identification mo-del based on TensorFlow-based deep CNN(Convolutional Neural Network) is proposed. The inspection image is pre-processed by image graying and size compression, and BM3D algorithm is used for image denoising to get the training data needed in the experiment. A deep CNN framework based on TensorFlow is proposed. By using the TensorBoard module in the framework, the model structure is designed and parameters of the deep CNN are selected, and the ReLU activation function and feature weight are analyzed theoretically. The experimental results show that after fifteen times of iteration training, the deep CNN has better recognition ability than the traditional SVM, ELM and BP neural network; compared with the classical LeNet-5 model, VGGNet model and models in the related literatures, the proposed model has more advantages. |
| Key words: power transmission lines foreign body identification convolutional neural network TensorFlow |
