Raspberry Pi 3 bundle for Kaggle dataset Wingbeats
We present a code bundle that allows wingbeat recordings to be transmitted through the Bluetooth wireless mode of a smartphone/tablet to a raspberry pi 3. The user sends the recording(s) and the result of the recognition taking place on pi3 appears on screen. We have embedded a CNN model on Python 2.7 and Python 3.5, Raspbian GNU/Linux 9 (stretch).
The weights are derived on a desktop computer using Keras 2.1.3 and after being converted to a Tensorflow 1.4 graph are ported on a pi3. The latter is only used to predict the correct class of the snippet.
The pi3 implementation suggests that a real-time wingbeat recognizer can be directly embedded to insect traps. Alternatively, a stand-alone pi3 can receive snippets through a Wifi functionality from a network of insect traps deployed anywhere in the world.
The following steps has been executed succesfully using android smartphones and iMac.
First, we need to make Pi able to receive files over bluetooth.
sudo apt-get install obexpushd
sudo nano /etc/systemd/system/dbus-org.bluez.service
Add the ' -C' flag to the end of the 'ExecStart=' line. It should look like this:
...
...
ExecStart=/usr/lib/bluetooth/bluetoothd -C
...
...
Save the file and reboot. Then create the bluetooth storage folder.
sudo mkdir /bluetooth
To run and automate the obexpushd service on boot, create this file:
sudo nano /etc/systemd/system/obexpush.service
and add the following:
[Unit]
Description=OBEX Push service
After=bluetooth.service
Requires=bluetooth.service
[Service]
ExecStart=/usr/bin/obexpushd -B -o /bluetooth -n
[Install]
WantedBy=multi-user.target
Save the file and set that to autostart with:
sudo systemctl enable obexpush
Reboot and press 'Make Discoverable' in bluetooth's setings to be able to find Pi and pair it with any device. If devices were already paired before the obexpushd service, re-pair them.
You should now be able to send files from your phone/tablet (or other device) to the Pi, they will appear in the /bluetooth directory. Notice that they will be owned by root, so you'll need sudo to access them.
run in terminal:
sudo apt-get install libblas-dev liblapack-dev python-dev libatlas-base-dev gfortran python-setuptools
sudo pip2 install http://ci.tensorflow.org/view/Nightly/job/nightly-pi/lastSuccessfulBuild/artifact/output-artifacts/tensorflow-1.5.0-cp27-none-any.whl
sudo apt-get install libffi-dev
sudo pip2 install pysoundfile
run in terminal:
sudo apt-get install libblas-dev liblapack-dev python3-dev libatlas-base-dev gfortran python3-setuptools
curl -o /tmp/tensorflow-1.5.0-cp34-none-any.whl -O http://ci.tensorflow.org/view/Nightly/job/nightly-pi-python3/lastSuccessfulBuild/artifact/output-artifacts/tensorflow-1.5.0-cp34-none-any.whl
mv /tmp/tensorflow-1.5.0-cp34-none-any.whl /tmp/tensorflow-1.5.0-cp35-none-any.whl
sudo pip3 install /tmp/tensorflow-1.5.0-cp35-none-any.whl
sudo rm /tmp/tensorflow-1.5.0-cp35-none-any.whl
sudo apt-get install libffi-dev
sudo pip3 install pysoundfile
Download and extract the wingbeats_pi-master.zip and cd inside wingbeats_pi-master folder. Choose samples from test_corpus folder and copy them into your device. You should be able to run the app anytime in terminal, with the following commands:
run in terminal:
python2 wingbeats.py
run in terminal:
python3 wingbeats.py