When countries want to measure quail numbers, the process can be cumbersome, time-consuming, and expensive.
It means spending hours in the field listening to calls. Or leave a recording device in the field to capture the sounds coming out – only to spend hours afterward listening to that sound. Then, repeat this process until enough information is available to start making population estimates.
But a new model developed by researchers at the University of Georgia aims to simplify this process. Using artificial intelligence to analyze terabytes of recordings of quail calls, the process gives wildlife managers the ability to gather the data they need in a matter of minutes.
“The model is very accurate, capturing between 80% and 100% of all calls on even the noisiest recordings. James Martin, an assistant professor at the UGA Warnell School of Forestry and Natural Resources who has been working on the project, in collaboration with the Georgia Department of Natural Resources, said for two years. Almost five years.” “This new model allows you to analyze terabytes of data in seconds, and what that will allow us to do is increase monitoring, so you can literally take out hundreds of these devices and cover a lot more space and do it with much less effort than in the past.”
The program represents nearly five years of the work of Martin, postdoctoral researcher Victoria Nolan, and several key contributors who worked with a code writer to create the model. It’s also part of a larger shift taking place in wildlife research, where computer algorithms are now helping work that once took humans thousands of hours to complete.
Increasingly, computers are getting smarter, for example, at identifying certain noises or features in images and audio recordings. For researchers like Martin, this means that hours once spent on tasks such as listening to audio or looking at game camera images can now be performed by a computer, freeing up valuable time to focus on other aspects of the project.
The new tool could also be a valuable resource for state and federal agencies looking for information on quail populations, but with limited funds to spend on any one project. “So, I think this is something that countries might jump on as much as they replace existing surveillance with audio recording devices,” Martin added.
The program’s success was recently documented by Journal of remote sensing in the environment and conservation.
Martin said that as the software becomes more used and exposed to voices coming from new geographies, it is getting “smarter.” As it is, quails offer several different types of calls. But when the software is exposed to a variety of sounds other than the quail, he said, it is better able to distinguish the correct calls from the ambient noise of the grass and trees around them.
Over time, the program will increase in recognition.
“That’s why you have to keep giving it the training data, and when you move the geographs, you encounter new voices that you haven’t trained the model on,” he added. “It’s always about adapting.”
Remote sensing of the environment and conservation
Development of a convolutional neural network for automatic detection of northern Bobwhite Colinus virginianus covey . calls
The date the article was published
August 24, 2022
Disclaimer: AAAS and EurekAlert! is not responsible for the accuracy of newsletters sent on EurekAlert! Through the contributing institutions or for the use of any information through the EurekAlert system.