Scientist develops AI photo traps for nature conservation

William Menz uses artificial intelligence to classify small mammals in order to research how the population of mice develops and thus support conservationists and biologists in their work. In his master's thesis, he trained neural networks to automatically recognize the animals.
The mice are lured into the photo trap with food. Afterwards, their images are evaluated with the help of neural networks.

Small mammals like mice are a real nuisance for farmers, because they often eat away their hard-earned harvest. But agriculture also harms the small animals - for example, when pesticides are used in the fields and the animals ingest them through their food. To get an overview of how harmful pesticides are for mice and co. and how the small mammal population develops on intensively farmed land, naturalists set traps there. In this way, the species that occur can be determined and counted. However, this method is very time-consuming because the traps have to be checked regularly to avoid stress for the animals. For the registration of shrews, the traps have to be checked every three hours. For this reason, they are rarely examined. Therefore, biologists and conservationists are increasingly using photo traps. Small mammals are lured into a small box with bait. There, a camera takes photos of them before they leave the trap on their own.

William Menz, an academic staff member at the Audiovisual Technology Group, has now added a new intelligent component to photo traps for small mammals. In his master's thesis "The classification of small mammals by comparing different neural networks, with the data collection through a camera trap", supervised by Prof. Alexander Raake and Dr. Eckhardt Schön at the Audiovisual Technology Group and in cooperation with the wildlife monitoring company Eurofins MITOX, he used neural networks to evaluate the images captured in the trap. With the help of the algorithms, which are modeled on the human brain in their basic structure, the photos were grouped according to certain characteristics such as the size of the ears or the shape of the animals' bodies and assigned to four classes. In this way, William Menz was able to identify whether the animals were forest mice, voles or shrews, or whether another animal had fallen into the trap.

Simplified evaluation of the photo traps

The methodology developed in the master's thesis greatly assists animal researchers in monitoring the small mammal population, as William Menz explains:

The effort required to view and analyze the images is significantly reduced. The AI can evaluate images in large numbers, so experts no longer have to be consulted to classify each image.

A total of ten photo traps were set up near Erfurt and in Lusatia over a period of several weeks. This resulted in tens of thousands of images, which were first sifted through manually by William Menz. The scientist selected a data set of several thousand photos, which he used to train the neural networks:

The networks were able to perform the basic tasks such as edge detection right from the start. Through training, the process was further refined. In the final layers of the network, more complex tasks were completed until the network was finally able to tell what was in an image. This determination is only ever possible for the task the network was trained to perform.

So far, AI has been used primarily for monitoring larger animals such as wild boar or wolves, and birds are also monitored using noise recognition for research purposes. The method for automated species identification using neural networks for small mammals such as rodents, mice or hedgehogs opens up new possibilities for research to classify the animals faster and more efficiently. For example, the company Eurofins MITOX wants to further use the photo traps and train the neural networks to such an extent that the exact species of an animal can also be recognized. In particular, the monitoring of shrews will benefit from this new approach.


William Menz

Academic Staff Member at the Audiovisual Technology Group