Artificial intelligence analysis of bird songs assisting researches analyse bird populations and movements
Research regarding bird songs carried out in the Sierra Nevada mountain ranges located in California produced about a million hours worth of audio output, which artificial intelligence researchers are working on decoding to obtain insights into how birds reacted to forest/wildfires in the area, and to further gain insight into what measures assisted the birds to bounce back in a proactive fashion.
Researchers can additionally leverage the soundscape to assist in tracking shifts with regards to migration timing and population ranges, going by a latest recounting in the Scientific American. Additional audio input is flowing in from other research work as well, with sound-based projects to count insects and research the impact of light and noise pollution on bird populations underway.
“Audio data is a treasure trove as it consists of massive amounts of data,” said ecologist Connor Wood, a postdoctoral researcher from Cornell University who is spearheading the Sierra Nevada project. “We are just required to contemplate creatively with regards to how to share and access that data.” AI is beneficial, and with the updated generation of machine learning, AI systems are capable of detecting animal species from their calls, and can undertake processing of thousands of hours worth of data in less than 24 hours.
Laurel Symes, assistant director of the Cornell Lab of Ornithology’s Centre for Conservation Bioacoustics, is researching acoustic communication in animals, which includes frogs, bats, crickets, and birds. She has undertaken compilation of several months of recordings of katydids (popularly vocal long-horned grasshoppers that are a fundamental part of the food web) in the rain forests of central Panama. Breeding activity patterns and seasonal population variation are obscured in this audio, but undertaking analysis is time-intensive.
“Machine learning has been the game changing factor for us,” Symes specified to Scientific American.
Symes and her colleagues logged 600 hours of work to categorize several katydid species from only 10 documented hours of sound. However, a machine-learning algorithm her team is producing, referred to as KatydID carried out the same task while its human developers “went out for a beer,” Symes specified.
BirdNET, a widespread avian-sound-recognition system available currently, will be harnessed by Wood’s team to undertake analysis of the Sierra Nevada recordings. BirdNET was constructed by Stefan Kahl, an ML scientist at Cornell’s Centre for Conservation Bioacoustics and Chemnitz University of Technology in Germany. Other scientists are leveraging BirdNET to record the impacts of light and noise pollution on bird songs during dawn in the French Briere Regional Natural Park.
Avian bird calls are complicated and varied. “You require a lot more than merely signatures to detect the species,” Kahl specified. Several birds have not just one song in their repertoire, and usually demonstrate regional “dialects” – a white-crowned sparrow from Washington State can sound really different in contrast to its Californian cousin, ML systems can isolate these variations. “Let’s assume there’s an yet unreleased Beatles song that is released today. You’ve never listened to the melody or the lyrics prior, but you are aware it’s a Beatles song as that’s what they sound like,” Kahl stated. This is exactly what these programs learn to do as well.
BirdVox Combines Study of Bird Songs and Music
Music recognition research is currently making the crossover into bird song research, with BirdVox, which is a collaborative effort between the Cornell Lab of Ornithology and NYU’s Music and Audio Research Lab. BirdVox intends to look into machine listening strategies for the automatic detection and classification of free-flying bird species from their vocalizations, going by a blog post at NYU.
The researchers responsible for BirdVox are hoping to undertake deployment of acoustic sensing devices for real-time monitoring of seasonal bird migratory patterns, specifically the determination of the accurate timing of passage for each species.
Present bird migration monitoring utilities are reliant on data from the weather surveillance radar, which furnishes insight into the density, direction, and pace of bird movements, however, not into the migration of species. Crowdsourced human observations are made nearly exclusively during daytime hours; they are of restricted use for researching nocturnal migratory flights, the researchers indicated.
Automatic bioacoustic analysis is perceived as being complementary to these strategies, that is scalable and capable of generating species-particular data. Such strategies have broad-ranging implications in the domain of ecology for comprehending biodiversity and monitoring migrating species in regions with buildings, planes, communication towers and wind turbines, the scientists observed.
Duke University Researchers Leveraging Drones To Monitor Seabird Colonies
In other news in bird research, a team from Duke University and the Wildlife Conservation Society (WCS) is leveraging drones and a deep learning algorithm to survey large colonies of seabirds. The team is undertaking analysis of more than 10,000 drone imagery of mixed colonies of seabirds in the Falkland Islands off of Argentina’s coast, going by a press release from Duke University.
The Falklands, also referred to as the Malvinas, are home to the planet’s biggest colonies of black-browed albatrosses and second biggest colonies of southern rockhopper penguins. Hundreds of thousands of birds breed on the islands in densely interlinked groups.
The deep-learning algorithm accurately detected and counted the albatrosses with 97% precision and the penguins with 87% precision, the team stated. Cumulatively, the automated counts were within 5% of human counts approximately 9/10ths of the time.
“Leveraging drone surveys and deep learning provides us an alternative that is very precise, less disruptive, and considerably easier. One individual, or a small team, can carry it out, and the gear you need to accomplish it isn’t expensive or complex,”specified Madeline C. Hayes, a remote sensing analyst at the Duke University Marine Lab, who spearheaded the research.
Prior to this new strategy being available, to monitor the colonies located on two rocky, uninhabited outer islands, teams of researchers would count the number of every species they could observe on an area of the island. They would extrapolate these numbers to obtain a population estimation for the entire colony. Counts usually required to be repeated for improved precision, a monotonous procedure, with the presence of researchers possibly disruptive to the breeding and parenting behaviour of the birds.
WCS researchers leveraged an off-the-shelf consumer drone to gather more than 10,000 individual photos. Hayes converted into a large-scale composite visual leveraging image-processing software. She then undertook analysis of the image leveraging a convolutional neural network, a variant of AI that deploys a deep-learning algorithm to undertake analysis of an image and differentiate and count the objects its observes – two differing species of birds in this scenario, albatrosses and penguins. The data was leveraged to develop comprehensive estimations of the cumulative number of birds identified in colonies.
“A CNN is loosely modelled on the human neural network, in that it learns from experience,” specified David W. Johnston, director of the Duke Marine Robotics and Remote Sensing Lab. “You go about training the computer to pick up on differing visual patterns, such as those made by black-browed albatrosses or southern rockhopper penguins in sample imagery, and with the passage of time it learns how to identify the objects shaping those patterns in other imagery like our composite photo.”
Johnston, who is additionally associate professor of the practice of marine conservation ecology at Duke’s Nicholas School of the Environment, stated that the emerging drone – and CNN-enabled strategy is broadly applicable” and massively enhances our capacity to monitor the size and health of seabird colonies globally, and the health of the marine ecosystems they inhabit.”