A Novel Time Series-to-Image Encoding Approach for Weather Phenomena Classification

TL;DR

This work targets identifying weather phenomena from received signal level time series in 4G/LTE networks by converting temporal data into images and framing the task as image classification with CNNs. The authors propose a novel time-series-to-image encoding that uses normalization, derivatives, and a polar-coordinate mapping to generate image sequences from fixed-length windows, enabling effective data augmentation and improved classification under limited data. They conduct extensive experiments to assess window length and augmentation strategies, finding longer windows and carefully designed augmentations improve convergence and accuracy, with the best setup achieving up to about 95% test accuracy on larger images. The study demonstrates the viability of signal-attenuation-based rainfall classification and suggests future work comparing with Gramian Angular Fields and Markov Transition Fields and extending to additional datasets and time-series tasks.

Abstract

Rainfall estimation through the analysis of its impact on electromagnetic waves has sparked increasing interest in the research community. Recent studies have delved into its effects on cellular network performance, demonstrating the potential to forecast rainfall levels based on electromagnetic wave attenuation during precipitations. This paper aims to solve the problem of identifying the nature of specific weather phenomena from the received signal level (RSL) in 4G/LTE mobile terminals. Specifically, utilizing time-series data representing RSL, we propose a novel approach to encode time series as images and model the task as an image classification problem, which we finally address using convolutional neural networks (CNNs). The main benefit of the abovementioned procedure is the opportunity to utilize various data augmentation techniques simultaneously. This encompasses applying traditional approaches, such as moving averages, to the time series and enhancing the generated images. We have investigated various image data augmentation methods to identify the most effective combination for this scenario. In the upcoming sections, we will introduce the task of rainfall estimation and conduct a comprehensive analysis of the dataset used. Subsequently, we will formally propose a new approach for converting time series into images. To conclude, the paper's final section will present and discuss the experiments conducted, providing the reader with a brief yet comprehensive overview of the results.

Figures (8)

• Figure 1: Plot of received signal's level under changeable wheater conditions (i.e., "sereno variabile").
• Figure 2: Plot of received signal's level under moderate rain conditions (i.e., "pioggia moderata").
• Figure 3: Plot of received signal's level under weak rain conditions (i.e., "pioggia debole").
• Figure 4: Random image samples obtained from the time-series-to-images procedure.
• Figure 5: Data distribution after applying the procedure without using data augmentation.