An Innovative Acoustic Rain Gauge Based on Convolutional Neural Networks

Abstract

An accurate estimate of rainfall levels is fundamental in numerous application scenarios: weather forecasting, climate models, design of hydraulic structures, precision agriculture, etc. An accurate estimate becomes essential to be able to warn of the imminent occurrence of a calamitous event and reduce the risk to human beings. Unfortunately, to date, traditional techniques for estimating rainfall levels present numerous critical issues. The algorithm applies the Convolution Neural Network (CNN) directly to the audio signal, using 3 s sliding windows with an offset of only 100 milliseconds. Therefore, by using low cost and low power hardware, the proposed algorithm allows implementing critical high rainfall event alerting mechanisms with short response times and low estimation errors. More specifically, this paper proposes a new approach to rainfall estimation based on the classification of different acoustic timbres that rain produces at different intensities and on CNN. The results obtained on seven classes ranging from “No rain” to “Cloudburst” indicate an average accuracy of 75%, which rises to 93% if the misclassifications of the adjacent classes are not considered. Some application contexts concern smart cities for which the integration of an audio sensor inside the luminaire of a street lamp is foreseen, precision agriculture, as well as highway safety, by minimizing the risks of aquaplaning.

A CNN-based Differential Image Processing Approach for Rainfall Classification

“…The operation of the labeling algorithm implemented within the Raspberry Pi, the connections between the hardware devices in use and the categories of recording intensity are explained in detail in [17]. The 4G modem key allows real-time data transmission [20].…”

“…In [6]- [7], the authors investigated the use of the radio frequency signal, used in the latest generation cellular systems, as a tool for classifying rainfall intensity levels using pattern recognition method. However, in [17] an acoustic rain gauge has been proposed which is able to classify the rainfall levels in 5 classes through rainfall timbre and deep learning techniques, in particular by applying convolutional neural networks [18], [19].…”

Assessment of Multimodal Rainfall Classification Systems Based on an Audio/Video Dataset

“…The processing unit allows the synchronous aggregation of audio and video sequences in the same precipitation intensity class. The labeling algorithm, described in detail in our previous paper [17], allows defining the precipitation intensity classes to which the audio and video files (each 22 seconds long) belong. The labeling takes place utilizing the "interruptions" generated by the tipping bucket type rain gauge, whenever it is overturned by rain.…”

“…Subsequently, the obtained sub-sequences are fed as input to the neural network. The result of this training and subsequent testing is described in detail in [17].…”

“…In particular, the database presents a set of audio sequences containing the acoustic timbre produced by the rain and video sequences containing rain videos, both in seven different intensities, i.e., "No rain," "Weak rain," "Moderate rain," "Heavy rain" and "Very heavy rain," "Shower rain" and "Cloudburst rain." The database proposed in this paper is an extension of the one used in our previous study [17], where we proposed a new rainfall classification algorithm based on an audio signal and evaluated its performance. At first, the spectral and statistical analysis of the audio sequences were presented, followed by the performances from the accuracy of an audio pattern recognition system based on the CNN network.…”