The new method improves the measurement of animal behavior using deep learning


The new method improves the measurement of animal behavior using deep learning

A deep learning tool, called DeepPoseKit, can automatically detect animal body parts directly with speed and accuracy from images or videos – without physical markers. This method for animals can be used in laboratory environments (eg owls and locusts) or in the wild (eg zebras). Credit: Jake Gravinger

The new toolkit goes beyond existing machine learning techniques by measuring body posture in animals at high speed and accuracy. DeepPoseKit, a deep learning tool developed by researchers at the Constance University Center for Advanced Study of Collective Behavior and Max Planck Institute for Animal Behavior, builds on previous approaches to positive assessment of modern achievement. In Computer Science. These newly developed deep learning techniques can accurately measure body posture after previously unheard images, after training only 100 examples, and can be used to study wildlife in difficult conditions. Published today in Open Access Journal eLifeThe research fosters the field of animal behavior with the next generation of tools, while at the same time providing the opportunity for non-experts to easily access machine learning for their behavioral research.


Animals need to interact with the physical world in order to survive and replicate, and studying their behavior may reveal solutions that have evolved to achieve these ultimate goals. Behavior is difficult to determine only by its direct observation: the bias of human observers and the limited processing power impede the quality and resolution of behavioral data that can be collected from animals.

Machine learning has changed. A variety of tools have been developed in recent years that allow researchers to automatically control animal body parts directly without pictures or videos – without the use of offensive markers on animals and manual behavior of animals. However, these methods have drawbacks that limit performance. "The means of measuring body posture through deep learning were slower and more accurate, or faster and less accurate. But we wanted to do the best in both worlds. " Says lead author Jake Gray, a graduate student at the Max Planck Institute for Animal Behavior.

Credit: Constance University

Credit: Constance University

In a new study, researchers have presented an approach that will overcome this trade-off in speed. These new methods use an effective, state-of-the-art deep learning model to image body parts to identify body parts, and a fast algorithm for calculating the location of these body parts with high accuracy. The results of this study also demonstrate that these new methods can be used in species and experimental conditions – from owls, locusts and mice in controlled laboratory environments to wildlife zebras. Dr Blair Costello, co-author of the study of zebrafish in Kenya, says: "The posture data that can be collected by zebra using DeepPoseKit allows us to know exactly what each individual in the group does and how they interact. Existing technologies such as GPS reduce this complexity. One point in space De, which limits the types of questions which you can answer. "

Because of its high-quality and easy-to-use software interface (the code is publicly available on Github, https://github.com/jgraving/deepposekit), researchers said DeepPoseKit could immediately benefit scientists in a wide variety of fields. Such as neuroscience, psychology and ecology, and levels of expertise. Work on this topic may also have applications that affect our daily lives, such as improving gesture-like algorithms on smartphones, or diagnosing and monitoring movement-related diseases in humans and animals.

"In just a few years, deep learning has ended up being a niche, one of the most democratic and widely used software tools in the world," said Yain Kucin, senior author for the paper. Who heads the Center for Advanced Study of Collective Behavior at Constance University and the Max Planck Institute for Animal Behavior Collective Behavior. "Our hope is that we can contribute to behavior research by developing easy-to-use and highly-developed tools that anyone can use." Such tools are important for studying behavior because, as Graevit said, "they allow us to start with the first principles, or" how The animal moves its body in space The e? ”Rather than subjective explanations of what behavior is. There, we can begin to apply mathematical models of data and develop general theories that will help us better understand how adaptive organization of individual and animal groups interacts.


Discovering the hidden intelligence of collectives


More info:
Jacob M Graving et al. DeepPoseKit, a software toolkit for deep and solid animal posture assessment using deep learning, eLife (2019). DOI: 10.7554 / eLife.47994

Magazine Info:
eLife

Provided by
University of Constance

Quote:
New Method Improves Measurement of Animal Behavior Using Deep Learning (2019, October 1)
Read October 1, 2019
https://phys.org/news/2019-10-method-animal-behavi-deep.html

This document is subject to copyright. For the purposes of private law and research other than a fair case, no
Part may be transmitted without written permission. Content is provided for informational purposes only.