光学学报, 2020, 40 (13): 1306001, 网络出版: 2020-07-09   

利用神经网络实现多模光纤传输散斑的识别 下载: 936次

Realization of Recognition for Multi-Mode Optical Fiber Transmission Speckle Using Neural Network
作者单位
北京交通大学光波技术研究所全光网络与现代通信网教育部重点实验室, 北京 100044
摘要
搭建实验平台,把26个字母的图像传入光纤,并在输出端采集散斑图。把散斑图展开到HSV色彩空间中,单使用V分量进行分类能达到不错的分类准确率,且能缩减训练时长。在预处理后,分别使用具有不同层数卷积结构的神经网络、卷积神经网络和支持向量机(CNN+SVM)算法、SVM算法对散斑图进行分类。测试结果发现,使用4420张散斑图作为训练集,3层卷积结构的神经网络的识别准确率为88%,4层卷积结构的神经网络的识别准确率为95%,CNN+SVM算法的识别准确率为98%,而SVM算法的识别准确率达到了100%。实验结果表明,把机器学习算法应用在光信号上,同样可以对多模光纤散斑图进行分类,当图像特征相对明显时,直接使用SVM算法对光纤输出散斑进行识别,可以大大提升多模光纤输出散斑图的识别准确率。
Abstract
Setting up an experimental platform, we transmit images of 26 letters to the optical fiber and collect the speckle patterns at the output end. When the speckle patterns are expanded into HSV color space, the single use of V component can achieve good classification accuracy and reduce training time for classifying speckle patterns. After pre-processing, the network with convolution of different layers, the convolutional neural network and support vector machine (CNN+SVM) algorithm, and SVM algorithm are used to classify speckle patterns. The test results show that using 4420 speckle patterns as a training set, recognition accuracy rate of neural network with convolution of three layers can reach 88%, recognition accuracy rate of neural network with convolution of four layers is 95%, recognition accuracy rate of CNN+SVM algorithm is 98%, and recognition accuracy rate of SVM algorithm can reach 100%. The experimental results show that some machine learning algorithms applying to optical signals can classify speckle patterns in multi-mode optical fiber. When the image characteristics are relatively obvious, direct use of SVM algorithm to identify the optical fiber output speckles can greatly improve the recognition accuracy of multi-mode optical fiber output speckle patterns.

卢顺, 谭中伟, 刘艳, 杨婧雅, 张利伟, 牛慧. 利用神经网络实现多模光纤传输散斑的识别[J]. 光学学报, 2020, 40(13): 1306001. Shun Lu, Zhongwei Tan, Yan Liu, Jingya Yang, Liwei Zhang, Hui Niu. Realization of Recognition for Multi-Mode Optical Fiber Transmission Speckle Using Neural Network[J]. Acta Optica Sinica, 2020, 40(13): 1306001.

本文已被 4 篇论文引用
被引统计数据来源于中国光学期刊网
引用该论文: TXT   |   EndNote

相关论文

加载中...

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!