Skip to content

Latest commit

 

History

History
103 lines (67 loc) · 5.01 KB

README.md

File metadata and controls

103 lines (67 loc) · 5.01 KB
Raw

Acoustic Rain Gauge

R & D project by ICFOSS

Introduction

Rainfall, also known as precipitation, is crucial for environmental stability. Accurate precipitation monitoring is vital for weather forecasting and creating early flood warning systems. In this project we have developed an acoustic rain gauge that estimates rainfall by using sound as input data.

Experiment Setup

Figure 1: Experiment setup

Installation and Setup

The setup guide can be found here.

Live Data and Visualization

grafana

Click here to see live data from our acoustic rain gauge

Components used

Sensor Used

  1. USB Mic, Jieli Technology UACDemoV1.0

Data Acquisition Devices (DAQs)

  1. Davis AeroCone 6466M Rain Gauge - The mechanical rain gauge used as a reference for comparing rainfall against acoustic readings.

  1. Raspberry Pi - Used as a DAQ device for high-resolution and high-sampling-rate audio recording.

Raspberry Pi

Data Collection

Data from USB Mic and Raspberry Pi

The USB mic captures audio files saved in WAV format with a fixed duration. The parameters for the audio files (e.g., sampling rate, sample duration, bit size, total recording time, etc.) can be set in the config.yaml file. The recorded audio files are analyzed for deep learning modeling.

The recorded WAV files are saved with a timestamp (yyyy_mm_dd_hh_mm_ss_millisec.wav) as shown below:

2023_11_06_16_13_11_011224.wav

Datasets Available on Kaggle

  1. Rain_Data_Master_2023 - Contains all audio recordings and rainfall data from the mechanical rain gauge collected so far. Note that files have different durations (10 sec and 3 sec) and sampling rates (48K & 8K).
  2. Rain_Data_Master_8K - Contains a downsampled version of Rain_Data_Master_2023 with a uniform sample rate of 8K.

Usage

  1. daq_pi.py and davis_logger.py - These are the main script files for data acquisition and model inference. daq_pi.py is used for automated audio recording on Raspberry Pi and model inference. davis_logger.py logs data from the Davis rain gauge using the Raspberry Pi's GPIO pins. These scripts can be initiated with the nohup command or by adding them to the ~/.bashrc profile. If added to the ~/.bashrc profile, the script will run everytime the device boots up/user logs in/terminal opened.
nohup home/pi/raingauge/src/daq_pi.py &
nohup /home/pi/raingauge/src/davis_logger.py &

OR

nano ~/.bashrc

# Appened the following line to the end of .bashrc file
python3 /home/pi/raingauge/src/daq_pi.py & python3 /home/pi/raingauge/src/davis_logger.py

# Reboot the device
sudo reboot

  1. mech_vs_non_mech_dataset_creation.ipynb contains the Kaggle script to combine wave files recorded along with mechanical rain gauge data to create training data for deep learning modeling.

  2. seq_mech_vs_non_mech.ipynb contains the LSTM modeling code which uses acoustic and mechanical data for rainfall estimation

Results

Epochs Model Features MAPE R² Correlation
25 LSTM STFT 15.76% 0.9683 0.9033
25 LSTM Chroma 45.15% 0.9700 0.1982
25 LSTM MFCC 96.53% 0.3418 0.4469

Table 1: Performance of LSTM model on various features

Team Members

  1. Gopika T G
  2. Sajil C K
  3. Manu Mohan M S
  4. Aiswarya Babu
  5. Harikrishnan K P

License

This project is licensed under the MIT License - see the LICENSE file for details.

Contact Information

You can find us at:

ICFOSS
Greenfield Stadium,
Opposite University of Kerala Campus, Karyavattom,
Thiruvananthapuram, Kerala 695581

For more information, visit our website.