Classification of Time-Series Images Using Deep Convolutional Neural Networks论文翻译
利用卷积神经网络对时间序列图像进行分类
Nima Hatami, Yann Gavet and Johan Debayle
Ecole Nationale Superieure des Mines de Saint-Etienne,
SPIN/LGF CNRS UMR 5307, 158 cours Fauriel, 42023 Saint-Etienne, France
ABSTRACT
Convolutional Neural Networks (CNN) has achieved a great success in image recognition task by automatically
learning a hierarchical feature representation from raw data. While the majority of Time-Series Classification
(TSC) literature is focused on 1D signals, this paper uses Recurrence Plots (RP) to transform time-series into
2D texture images and then take advantage of the deep CNN classifier. Image representation of time-series
introduces different feature types that are not available for 1D signals, and therefore TSC can be treated as
texture image recognition task. CNN model also allows learning different levels of representations together with
a classifier, jointly and automatically. Therefore, using RP and CNN in a unified framework is expected to boost
the recognition rate of TSC. Experimental results on the UCR time-series classification archive demonstrate
competitive accuracy of the proposed approach, compared not only to the existing deep architectures, but also
to the state-of-the art TSC algorithms.
卷积神经网络（CNN）通过自动化在图像识别任务中取得了巨大成功
从原始数据中学习分层特征表示。而大多数时间序列分类
（TSC）文献主要关注一维信号，本文采用递归图（RP）将时间序列转换为
2D纹理图像然后利用深度CNN分类器。时间序列的图像表示
介绍了不适用于1D信号的不同特征类型，因此可以将TSC视为
纹理图像识别任务。CNN模型还允许学习不同级别的表示
一个分类，联合和自动。因此，预计在统一框架中使用RP和CNN会有所提升
TSC的识别率。UCR时间序列分类存档的实验结果证明
提出的方法的竞争准确性，不仅与现有的深层架构相比，而且还与之相比
最先进的TSC算法。
Keywords:
Convolutional Neural Networks (CNN), Time-Series Classification (TSC), Deep Learning, Recur-
rence Plots (RP)
关键词：卷积神经网络（CNN），时间序列分类（TSC），深度学习，重复
rence Plots（RP）
1. INTRODUCTION
A time-series is a sequence of data points (measurements) which has a natural temporal ordering. Many important
real-world pattern recognition tasks deal with time-series analysis. Biomedical signals (e.g. EEG and ECG),
financial data (e.g. stock market and currency exchange rates), industrial devices (e.g. gas sensors and laser
excitation), biometrics (e.g. voice, signature and gesture), video processing, music mining, forecasting and
weather are examples of application domains with time-series nature.1–3 The time-series analysis motivations
and tasks are mainly divided into curve fitting, function approximation, prediction and forecasting, segmentation,
classification and clustering categories. In a univariate time-series classification, xn → yn so that n-th series of
length l: xn = (x1n, x2n, ..xnl ) is associated with a class label yn ∈ {1, 2, .., c}. It is worth noting that although
this paper is mainly focused on time-series classification problem, the proposed method can be easily adapted to
the other tasks such as clustering and anomaly detection.
时间序列是具有自然时间排序的数据点（测量）序列。很重要
现实模式识别任务处理时间序列分析。生物医学信号（例如EEG和ECG），
财务数据（如股票市场和货币汇率），工业设备（如气体传感器和激光）
激励），生物识别（例如语音，签名和手势），视频处理，音乐挖掘，预测和
天气是具有时间序列性质的应用领域的示例。1-3时间序列分析动机
任务主要分为曲线拟合，函数逼近，预测和预测，分割，
分类和聚类类别。在单变量时间序列分类中，x n → y n使得第n 个系列
长度升：X Ñ =（X 1 Ñ，X 2 Ñ，..x Ñ 升）与类标签相关联ÿ Ñ ∈{ 1 ，2 ，...，C } 。这是值得一提的是，虽然
本文主要关注时间序列分类问题，该方法可以很容易地适应
其他任务，如聚类和异常检测。