摘要
针对由于人为及外界复杂环境因素的干扰,接触网可视化接地系统的视频图像识别率较低的问题,提出了一种经结构简化与改进的VGG—16深度网络模型。通过迁移学习策略获得了良好的特征提取能力,结合稀疏自编码器(SAE)优化卷积特征,使网络具有一定的稀疏性,提升模型识别精度和泛化能力,又使训练参数减少。实验结果表明:该模型在验证集上达到了92%的识别率。
Aiming at the problem of low recognition rate of video image of visual over contact system( OCS)earthing system of connection network due to the interference of human and complex environmental factors,a simplified and improved VGG—16 model for depth network is proposed. Benefit from the transfer learning strategy,good feature extraction ability is obtained,and convolution feature is optimized by sparse automatic encoder( SAE),which makes the network sparse,it also improves the recognition precsion and generalization ability,reduces training parameters. The model achieves recognition rate of 92 % on validation set.
作者
吉鑫
陈剑云
完颜幸幸
JI Xin;CHEN Jianyun;WANYAN Xingxing(School of Electrical Engineering and Automation,East China Jiaotong University,Nanchang 330013,China)
出处
《传感器与微系统》
CSCD
北大核心
2021年第2期58-60,72,共4页
Transducer and Microsystem Technologies
基金
国家自然科学基金资助项目(51467004)。
关键词
可视化接地
图像识别
迁移学习
稀疏自编码器
visual over contact system(OCS)earthing
image recognition
transfer learning
sparse automatic encoder(SAE)
