Sensor Classification Using Convolutional Neural Network by Encoding Multivariate Time Series as Two-Dimensional...

Area of Science:

• Machine Learning
• Data Science
• Signal Processing

Background:

• Multivariate time series data is prevalent in sensor applications.
• Effective classification of sensor data is crucial for various industries.
• Existing methods may not fully leverage the spatial-temporal patterns within sensor data.

Purpose of the Study:

• To propose a novel framework for sensor classification using multivariate time series data.
• To evaluate the effectiveness of image transformation techniques for time series data encoding.
• To assess the impact of Convolutional Neural Network (ConvNet) architecture complexity on classification performance.

Main Methods:

• Encoding multivariate time series data into 2D images using Gramian Angular Summation Field (GASF), Gramian Angular Difference Field (GADF), and Markov Transition Field (MTF).
• Concatenating transformed images into a single larger image.
• Classifying the concatenated images using Convolutional Neural Networks (ConvNet).
• Evaluating performance on two open multivariate datasets.

Main Results:

• The choice of image transformation method and concatenation sequence did not significantly impact classification accuracy.
• A simple ConvNet architecture achieved performance comparable to complex architectures like VGGNet.
• The proposed image-based ConvNet framework demonstrated superior classification accuracy compared to other existing methods.

Conclusions:

• The proposed framework offers an effective approach for sensor classification from multivariate time series data.
• Image transformation techniques combined with ConvNets provide a robust method for analyzing sensor data.
• Simple ConvNet architectures are sufficient for achieving high accuracy in this classification task.