Deep learning tool improves chicken welfare by identifying distress calls
The deep learning model developed by the CityU team can identify and quantify chicken distress calls with 97% accuracy.
The deep learning model developed by the CityU team can identify and quantify chicken distress calls with 97% accuracy.

 

A research team led by City University of Hong Kong (CityU) has developed a deep learning model that can identify and quantify chicken distress calls from natural barn sounds with 97% accuracy. This breakthrough will help improve conditions and the welfare of chickens raised on crowded commercial farms.

The research is led by Dr Alan McElligott, Associate Professor, and Dr Liu Kai, Assistant Professor, in the Department of Infectious Diseases and Public Health at the Jockey Club College of Veterinary Medicine and Life Sciences at CityU, in collaboration with Imperial College London, Queen Mary University of London, the University of Surrey and the Guangxi Veterinary Research Institute. Other members include Ms Mao Axiu, PhD student, and Ms Claire Giraudet, Research Assistant, in CityU’s Department of Infectious Diseases and Public Health.

The annual global production of chickens exceeds 25 billion birds, which are often housed in very large groups, numbering thousands. Distress calling triggered by various sources has been suggested as an ‘iceberg indicator’ of chicken welfare, which can indicate mortality and growth rates. However, to date, the process of assessing distress calls has relied largely on manual annotations, which is labour intensive, time-consuming, and prone to the subjective judgements of individuals.

The research team collected and analysed recordings of ‘spotted’ and ‘three-yellow’ breeds in a poultry farm in Guangxi, with approximately 2,000 to 2,500 birds per house, and developed a new automated, objective and cost-effective method of assessing and quantifying distress calls, based on deep learning combined with bio-acoustic techniques.

The algorithm accurately detects when the chickens are stressed owing to their internal physical condition or external factors.
The algorithm accurately detects when the chickens are stressed owing to their internal physical condition or external factors.

 

The algorithm covers frequency ranges from 0 Hz to the Nyquist frequency of 11,025 Hz, which allows distress calls to be distinguished from natural sounds in the barn with 97% accuracy and accurately detects when the chickens are stressed owing to their internal physical condition or external factors, like overcrowding, not getting enough food and water, or attacks from other chickens.

“Sometimes it’s difficult to convince farmers who have to deal with producing these animals at a set price for supermarkets and everyone else to adopt technology to improve their welfare,” said Dr McElligott. “Our end goal is not just to count distress calls, but to create conditions in which the chickens can live with less distress.”

“In the future, this technology will potentially allow staff to monitor chicken welfare in real time and remotely, promoting earlier husbandry interventions when necessary. This can also reduce the workload of analysts and facilitate the analysis of large datasets, thus improving the husbandry and management of the animals,” said Dr Liu.

“Our algorithm fully considered the constraints in computation resources and is suitable for practical deployment on farms,” said Ms Mao.

The paper was published in the Journal of the Royal Society Interface, and the team expects the technology to be deployed commercially within five years.