On November 4 2018, the Cacophony Project was privileged to become a signatory of the Banks Peninsula 2050 Predator Free Initiative. Along with key personnel from DOC, Ecan, Christchurch City Council, Banks Peninsula Iwi and Banks Peninsula Conservation Trust, the Cacophony Project's Clare McClennan was present to participate in the signing ceremony held at the Living Springs amphitheater. This historic partnership aims to focus and coordinate efforts to eliminate invasive predators on Banks Peninsula.
This article comes to us from Tim Armitage who has recently installed a Cacophonometer at his property to help monitor the effects of predator controls.
Early October 2018 we received our Cacophonometer at our Sandspit home. The setup process was very simple with the app having been pre-installed and the main next steps being getting our account created online and the device registered. The location we chose was around 20 meters from our house (a site with power available) and a small WiFi extender soon fulfilled the coverage required to ensure the upload process was reliable. We soon had plenty of recordings to sample with the ‘meter following the pattern set by the software – i.e. greater intensity of records around dawn and dusk.
The thermal video footage from the Cacophonator devices has proven to be invaluable for machine learning and studying predator behaviour. Occasionally, however, we noticed that recording for some animals was starting later than it should. In this article we discuss the reasons why this was happening and what we have done to improve our animal detection algorithm.
David Blake is a semi-retired property investor who now likes to spend his time trapping pests and planting native trees. From time to time he helps out by volunteering at The Cacophony Project doing some filming and occasionally contributing to parts of the software.
The video in this link shows a visual representation of the contributions to the software components of the project over time. The visualisation is based on source code changes pushed to Github and shows the areas of the software being worked on and the people involved. You can see how the project builds momentum as more and more clever people get involved. It also highlights the sophistication of the software that links all the parts of our predator-free projects together.
The Cacophony Project started with the development of a simple cell phone based tool that can measure birdsong so there is an objective measure of how well birds are doing as predator control is rolled out. This product is now ready to be more widely tested around New Zealand.
The last couple of months have seen us make more improvements to our Cacophonator hardware. The changes made have been driven by the demands of upcoming projects and targets.
Up until now we've typically run our devices on mains power. We've preferred sites which are near native bush but have access to a wall socket (sometimes with long extension cords!). This has gotten us quite far in terms of testing our prototypes and gathering footage to train our machine learning classifier but obviously isn't going to be a long term solution. Being able to run on battery power opens up a huge range of new areas to our devices.
The Cacophonator hardware now incorporates a buck-boost converter which allows it to work from a number of types of power sources including various battery technologies with differing output voltages (which change as the battery discharges). The buck-boost converter also continues to support mains power using a classic "wall wart" AC adapter.
We explored a number of options for battery power and after a number of false starts and experiments we've found a New Zealand based manufacturer who will make weatherproof lithium-ion battery packs which meet our needs exactly. These packs have performed well in the cold and in heavy rain. A Cacophonator can run for 5-6 nights (turning off during the day) on a single battery pack and we have ideas on how to extend battery life further.
Acoustic recordings of birds have been used by conservationists and ecologists to determine population density of specific bird species in a region. However, it is very hard to analyse and visualise the presence/absence of a specific bird species by manually hearing these recordings even by an expert bird song specialist. I am working on developing computational tools to automatically classify and visualise bird sounds in order to recognise different bird species in the wild. It is a powerful combination of machine learning, ecology, and applications of multimedia visualisation. These tools can be used by conversationalists, ornithologists, ecologists, and evolutionary scientists to visually identify a bird’s species using their sound alone.
In this blog, I want to share some of the initial results obtained by automatic clustering of bird species based on their sound features using machine learning techniques. It is an attempt to find similarities between sounds of the same bird species and with other bird species as well as differentiating between birds and human sounds. Many thanks to Nirosha Priyadarshani, Stephen Marsland, Isabel Castro, and Amal Punchihewa for making their datasets available for further research. These datasets are available at https://github.com/smarsland/AviaNZ.
The Cacophony Project has very long-term goals to enable us to eliminate 100% of predators. It’s easy to look at what we are doing and assume that this will never work in the depths of the remote bush. While we acknowledge that there are many steps before we achieve that capability, the tools developed while getting to that end goal will be useful for other important parts of the problem.
Hi I’m Clare and I’m the newest member of the core Cacophony Project team.
I’m a passionate outdoor adventurer, native flora and fauna lover and an experienced software developer. I love the interesting and diverse nature of the Cacophony Project and find it exciting to be able to work on a project aligns perfectly with my values and interests.
At first look it seems like it should be easy to work out how effective traps are. Our first way of measuring it was very simple - how often do we see animals around a trap compared to how often they are caught by the trap. This is too simplistic: with enough time an animal will eventually wander into a trap and be caught. Any device with some chance of killing/luring will have a 100% success rate given infinite time. Therefore, time should be one of the parameters for a device’s effectiveness.
You have to love the way journalists make a headline. They get most of the details of the actual project pretty right though
Below is a talk from the recent New Zealand AI conference. It gives an up-to-date summary of the project with particular reference to how we are using Machine Learning.
As with all projects when the fun development parts are proven there is refinement required to make the devices more robust and usable. Below are a few photos of the next iteration of our hardware that makes the product more reliable and flexible.
We have been invited to present our work at New Zealand's AI conference. There is currently huge interest in exploring applications for AI around the world and in New Zealand. The Cacophony Project is a great example of how this field can be used to tackle a very New Zealand specific problem.
To book your tickets to the event click here.
The first step required to become 100% predator-free is to know for sure what type of predators are out there. The most common methods used to do this are tracking tunnels and chew cards. These tools require significant manual work and miss a lot of predators. Tracking cameras do a much better job of tracking predators and detect 2-10 times more than tracking tunnels.
Latest version of the Cacophonometer can be downloaded from link below. In this version:
With the generous help of Willowbank Wildlife Reserve we have been collecting thermal video footage of kiwis to help train our machine learning based animal classifier.
Here's a sample of the recordings we've been collecting...
The Machine Learning (artificial intelligence) identification has two main parts. A tracker locates any objects of interest and separates them from the background. These tracks are then passed to the classifier which analyses them to identify what sort of animal they contain.
Previously we used a static camera so the tracking part of the problem was quite easy as only the animals moved, whereas background objects (such as trees) stayed relatively still. This is obviously harder when the camera is moving. Below shows some examples of the new tracker working with the moving camera.
In previous blog post we have talked about the advantages of having a camera that can track animals. It will obviously allow better identification but also the ability to eliminate with something like a poison paintball.
The video below shows the latest experiments. We are just using a radio controlled car for testing but you will get the idea. It is a little amusing to see the artificial intelligence try to guess what the radio controlled car is.
We now have many thermal cameras deployed at various locations collecting recordings every night. This means we have a lot of thermal video footage to manually tag every day so that they can be used to improve our machine learning classifier. As I write this, we have collected almost 30,000 thermal video recordings.
Our recordings include many false positives - where a non-animal object triggered our camera's motion detector. We also have many, many recordings of birds - particularly at dawn and dusk. Filtering through all the false positives and bird footage is very time consuming.
This is just a quick peek at some very encouraging results showing that Artificial Intelligence can successfully work out what it is seeing from the video. We will put up more discussion and details next year.
The raw video is on the left. On the right, the animal is identified and the cumulative classification of the animal is at the top. The instantaneous guess from the Artificial Intelligence is changing in real time at the bottom.
Video 1: Example of classification of animals using AI
Hi everyone, my name is Matthew, I have been brought on to help with the machine learning side of things. I’m very excited to be part of this project.
My job here is to take all the thermal footage we have been recording and identify the animals in it.
It has been shown that trail cameras are much better at detecting predators compared to tracking tunnels, which are the default detection tool. Trail cameras are between 2-10 times more sensitive at detecting predators than tracking tunnels depending on species. This is obviously very important as we need to be able to measure all predators that are out there.
Hi everyone, I’m Finn. I’ve been involved with Cacophony for a year now. I started out on work based around the Cacophonometers from a hardware and business model angle. Most recently I’ve been working on software and the development of a method to analyse all of this birdsong we’re getting! This post will describe the work I’ve done and basically what The Cacophony Index 1.0 is.
In previous work we thought we had the ultimate predator detection camera. But our goal is to detect all predators so we have chosen to develop a higher resolution heat camera. Here's a reminder of why we want to detect all predators: