We will be using this thread to post updates from the project. In this post, we are sharing the next steps for the projects and the things we need to tackle in the next 3 months. We would love your participation and hope you will be interested in following our steps.
Objectives
Our aims for the next phase of this project are:
To classify bats’ calls on a Raspberry Pi.
To optimise for low-power consumption and long-range transmission.
To document the project: the enclosure, the assembly, the installation, and the deployment.
Roadmap
Migration to RPi
Migrate Bat Classifier Algorithm to RPi
Rework data transmission
Rework database storage
The advancement in the project will happen mainly on the GitHub Repository Shazam4BB. The repo is currently being built, but you can expect the following by the end of the project:
Documentation Folder: (1) design files to build enclosure, (2) manual for assembly, (3) manual for installation, (4) manual for operating the device.
Scripts Folder: (1) script to record, process, and analyse audio files, (2) script to transfer data and access data.
If you are interested in the current “Shazam for Bat” device, you have a look at this blogpost.
Contact
We are looking forward to keeping you posted on our progress and look forward to hearing your ideas, comments, and remarks.
Please feel free to directly comment on this post, write to @audevuilli or drop us an issue in the GitHub repository.
Thank you to GOSH and the GOSH Community for all your help and support.
Hi Everyone,
We would like to update you on the Shazam for Bats project.
Activities Summaries
Feb-Apr 23 - Develop Software Application to run BatDetect2 deep-learning model on RaspberryPi 4B.
Apr 23 - Deploy and test application over a week.
Apr 23 - Make BOM and purchase items to test 3 long-term deployments in the wild.
Mai 23 - Present version 1 of the system at the BritBats23 Conference.
Project State
After discussing how to develop a bat classifier for single-board computers, we decided to take a step back and not only develop an application for bats but an application that could easily be modified to be used in other acoustic classification tasks.
We set up the following design requirement to sketch and develop such an application: (1) recording parameters such as recording duration and intervals are easily modifiable, (2) different types of ML acoustic models can be implemented, (3) options are available for saving and deleting recordings and detections files.
Below is the process diagram that summarises the first version of the application that implements the BatDetect2 algorithm developed by Oisin Mac Aodha and Santiago Martinez to detect bat calls and classify bat species.
The application has been deployed and tested for a week. The results are promising. The next step is to deploy 3 devices in the Queen Elizabeth Olympic Park (QEOP) in London with an updated version of the application. The updated version sends the bat calls detection and classification directly to a remote server via MQTT.
Next Steps
Finalise documentation for the software application.
Finalise documentation for deploying devices.
Put together the 3 devices to be deployed.
Notes
Friday 5th May, the BritBats Conference is taking place in Bristol. We will discuss the BatDetect2 model and present the first version of Shazam4Bats.
We are currently developing the application in a closed GitHub repository but are looking to open it to the public once we have finalised the documentation and run a few more tests. Hopefully, in the next few weeks We will keep you posted on the release.
We will soon be looking for people to test the implementation. So keep an eye open!
I’d be interested in testing the implementation as I’ve been thinking of trying to run BatDetect2 on a Raspberry Pi for my local bat group so am delighted that this project already exists.
Is there a timescale for a release of the application? I’m a Software Engineer (mainly in Python) and am keen to help try it out on my Raspberry Pi
For anyone who is interested in bats. We have identified the main roosts in our building as pipistrelle soprano. They are tiny little cuties, they roost in tiny spaces underneath the roof slates. Unfortunately a few died this year due to poor hunting conditions with all the rain, we worked with some local experts to care for the ones that didn’t make it back to the roost at night.
However an examination of the bat poo in other areas of the building we think the site is a temporary roost or hunting roost for lesser horseshoe and grey long eared bats. These roosts are inside the main building. They really like to poo all over one part of the machinery.
@hpy and I have been talking about putting in a camera trap, audio detection too would be really cool.
@julianstirling how cool! Yeah I’m definitely still up for some camera-trapping and have camera traps ready to go. Are Saturdays the better time to do this? Hope to see you soon and do some field work!
Thanks for the reply. They are so cute, unfortunately if they are to weak to reach the nest then their survival rate is depressingly low. To help them we have to pick them up (gloved) and inspect their wings, and try to provide them hydration and a safe place for until evening. In the evening we can try to help set them on the stone wall to climb to their roost. We had never had one die on us until this year, but we lost about 10 this year. The good thing is we do have 100s in the roof, so it is great to know the colony as a whole is still thriving. I just hope next year the weather isn’t so antagonistic to their hunting.
The suggestion of an AudioMoth is something I have considered, and I should probably just bite the bullet and get one. Currently the charity that runs the mill has other priorities/demands; wit the increased storms recently we have flooded 19 times since October 2022. A flood in January flooded not just the ground floor, but came all the way up the stairs and flooded the next floor. Repeated flooding takes a toll on the building not to mention the machinery inside it. Unfortunately the flooding is also limiting out opening season to the public to bring in funds. So while at £120 or so the AudioMoth is super affordable comparatively to other options, it is a big outgoing for an organisation on the edge.
Probably an answer to this is I buy one for myself and put it at the pump. Hopefully my work will pick up soon and I will do that.
Either way, soon @hpy and I will get some camera traps sorted. And once the bats are out of hibernation we can fill the GOSH forum with pictures of healthy bats flying!
Sorry to hear you lost some but it’s fantastic that you’re helping them alongside the commitments at the mill (I found the site via your profile). Fingers crossed this year is kinder!
Understandable that the charity has other priorities, is there a local bat group that could help survey the site?
We will keep you posted on the release.
