Latest progress in Smart Life Traps: more catches and detailing

This project aims to develop smart life traps that use AI image-recognition. By incorporating AI we prevent unwanted bycatches of protected species as European beaver or otter and only catch target species like coypu and muskrat. More catches are made in the field and more details on the traps are improving.

By now, ten smart life traps are in use in Germany, five in Belgium and ten in the Netherlands. Two coypu and six muskrats were caught on either side of the Dutch-German border. In the past months AI image-recognition software was improved by using field images from the smart life traps. Also, the module works now both on- and offline, even when it is not connected to the internet.


On hardware, the traps are improved as well. Mesh is mounted on both sides of the cage and the ground plate is elevated in order to obtain better pictures from animals that enter the trap. Also, mesh is mounted in front of the camera system for protection.

This summer the hardware will further develop with improvements like a stronger magnet and battery. Also, the AI image-recognition software will be enriched with images of otters and raccoons. The online message sent to the trapper will soon improve by always sending correct information on battery, GPS and trap status.

Active smart life trap at Sint-Maartensheide – De Luysen, April 1th 2022.

The message that is send to the trapper.

First results on survey methods and materials

A survey was made on methods and materials for eDNA, DNA sequencing/mapping, smart cameras and smart traps. Although responses were few, preliminary results include excellent scores for DNA methods and useful field results for smart systems.

A total of 21 responses was too few for statistical analysis. This was the first survey, so possible effects may be analysed over time.

However, the first results include excellence in protocols for handling both DNA methods. It looks like eDNA is an excellent warning system and DNA sequencing/mapping is extremely effective for determining migration routes.


Challenges are the intensive maintenance of materials (eDNA and DNA sequencing/mapping), the slow process (DNA sequencing/mapping) and the contradictory in work: trappers understandably prefer to work in areas with dense populations of rats, whereas eDNA methods focus on areas with few muskrats (eDNA).

Smart systems

For the smart systems (cameras and traps) the first results show that they are both multifunctional and can be used for different species. Smart life traps tend to render better catches. Design of the smart cameras is considered excellent.


Improvement on the traps is needed on the hardware (battery and magnet), software (AI image-recognition) and the message to trappers (battery level, GPS and trap status). On the smart cameras there is room for improvement on the AI image-recognition software used for detecting muskrat or coypu and on the speed of processing.

A new survey will be held in October.

DNA Mapping in Friesland is going well

In January 2022, the results of the first round of DNA mapping in Friesland were presented to the muskrat fighters. Based on the results, the catch intensity along the probable inflow locations in Friesland will be increased by placing a cordon of traps here. Smart wildlife cameras have been installed at the intake locations.

Red cross: influx from outside Friesland
Yellow cross: transfer in Friesland

Tail tips

In addition, in the period from February 2022 – February 2023, the tail tips of all muskrats caught in Friesland will be collected for follow-up research. The aim is to collect 100 muskrats this year, spread as much as possible across Friesland. More than 50 muskrats have already been collected.

eDNA subproject is scaling up

The current aim for the muskrat eDNA part of Life-MICA is scaling-up and real-life implementation of the eDNA approach and transference of sample processing to the water laboratories.

In order to compare results between the lab at the University of Amsterdam and the labs at Wetterskip Fryslân and Waterproef (Noord-Holland) we started sampling this year in regions with a higher chance of muskrat presence. For Fryslân this meant sampling near the borders, and for Noord-Holland sampling in a region where they have trapped the most muskrats in the past few years. Both regions have a low population of muskrat compared to other parts of the Netherlands.


In Fryslân a polder that has been designated as empty of muskrat was also sampled to determine if the eDNA results matched the evaluation based on traditional tracking methods (polder empty of muskrat), this did indeed match (Fig 1, region indicated by blue arrow). Figure one shows the results of sampling in Fryslân.

As of 29-06-22, 449 monitoring samples haven been taken in Fryslân, most by boat, as well as 78 localisation tracks, 70 point samples and 9 control samples (taken to confirm no more muskrat are present after catch actions).

As of 22-06-22, 34 muskrats had been trapped in the sampled areas (27 in waterways/tracks with eDNA and 7 in tracks without eDNA, but that were adjacent to tracks with eDNA).


In North Holland, there were many more traces of muskrat eDNA in the sampled region than in previously sampled areas. This allowed us to make a good comparison between the laboratories. But the high number of samples containing muskrat eDNA makes this region less suitable for the mainstream eDNA approach. Because of the significant population of muskrat in the area in the period of May/June the trappers focussed on catching the muskrats in this area. Figure 2 shows the catches of the muskrat overlayed on the eDNA results.

As of 22-06-22, 215 muskrat had been caught in the sampled area (182 in waterways/tracks with eDNA and 33 in tracks without eDNA, but again these were adjacent to tracks with eDNA). In Noord-Holland 167 monitoring tracks were sampled, as well as 163 localisation tracks and 181 point samples.

Most tracks in Noord-Holland were sampled by hand, which is more labour intensive and time consuming than sampling by boat. Areas with these population levels are more suited for follow up with eDNA after an intensive trapping effort, to determine if there are remaining muskrats, rather than the method used for monitoring areas that are historically empty or have very low presence.


For the Coypu the aim is not so much scaling up but determining how to best integrate the eDNA method in the tracking efforts. Coypu behave differently from muskrat and are also caught alive in cages instead of traps in burrows. This makes certain parts of the field approach less useful for coypu (localisation of burrows). 31 areas were sampled for coypu presence, and as of 22-06-22, there were 23 catches, of which 21 corresponded with eDNA signal.

Wetterskip Fryslân and Waterproef processed all the samples of their respective regions for this year, and are thus capable of routinely processing samples. Transfer of analysis of coypu eDNA samples has been initiated with Aqualysis.

Figure 1. Results monitoring Fryslân. Green: eDNA negative, Yellow: eDNA weakly positive and Red: eDNA positive. Both yellow and red tracks are followed up in the protocol.
The blue arrow indicates the polder which was sampled in order to determine if there were muskrat remaining in an area that had been marked as empty by traditional methods. This polder was sampled by hand/quad.
Figure 2. Sampled area Noord-Holland
Green: eDNA negative, Yellow: eDNA weakly positive and Red: eDNA positive.
Blue triangles: catches

Mallard the most photographed with the camera trap

In the different project areas 47 camera traps have been placed to detect muskrat and coypu presence. These cameras take a sequence of images when they are triggered by movement.

Afterwards trappers need to annotate these sequences to see which species is on there. Over 80.000 sequences have been annotated so far. The most commonly seen species on the cameras are mallards.

Artificial Intelligence

Since June of 2021 we have been using the Artificial Intelligence (AI) developed by the Agouti team to help annotate our images and ease the workload. We started a project with the team from Agouti to retrain their AI to better work for water conditions. Hopefully this will further optimize the workflow for detecting muskrats and coypu with camera traps.

Figure 1: distribution of the species observed on the camera’s.
Figure 2: muskrat caught on camera

A further catch with the ‘intelligent live traps’

Intelligent live traps have been set up at the Aschauteiche ponds to further test and develop the AI module of the live traps under real field conditions. On 6 April 2022, the first coypu was caught in the German project areas. The progress in development here has shown that the AI correctly detects coypu and other species and non-target species are not caught.

On the night of 6 April, we had the first capture of a coypu in the project area Aschauteiche. The intelligent live trap, equipped with the AI module with the recognition software, detected a coypu as it entered the trap, activated the closing mechanism, and then sent a message about the catch, including pictures, to the trap supervisors.

Intelligent live trap at the ash ponds

A total of 10 intelligent live traps are currently set up in Lower Saxony, including at the Aschauteiche, an area where not only the invasive species coypu occurs, but also to raccoons and common raccoon dogs, as well as protected species such as the otter.

The image recognition software installed in the back of the traps was able to document rats, mice and raccoons in the trap. Here, the AI module is being further trained to improve the system.

Wildlife cameras that monitor the trap prototypes from the outside and document the presence of animals with the trap provide additional certainty. This allows us to evaluate the functionality of the prototype more precisely.

Below, you can find some pictures of the coypu capture, as well as photos of the raccoons’ nightly ‘forays’ into the traps.

The coypu observed from the photo traps that supervise the intelligent cages.
The KI systems collects photos inside the cage to evaluate which animal should be caught and which are the non-target species.
As soon the KI system has taken the decision to close the trap and catch the coypu, Telegram informs the trappers sending a message with the necessary information and pictures of the catch. Thereafter, the coypu is took out from the catch.
Here an example of a raccoon going out of the cage, which was correctly not caught.
The same raccoon as above, documented by the image recognition software of the trap.

‘Smart life trap’ recognizes and catches first coypu on Dutch border

For the first time a coypu has been recognized and caught by a ‘smart life trap’, rigged with image-recognition on the Dutch-German border near Winterswijk. This animal species is not native to Europe and this new technology will help water authorities to catch them.

A ‘smart life trap’ uses image-recognition to recognize and trap animal species: the cage closes catching muskrats and coypu, while it remains open when entered by, for example, birds or otters. In January a coypu in the Wooldseveen moor near Winterswijk walked into the cage, was recognized and caught. It was the first catch in the field, after a long preliminary phase of testing and refining.

More efficiënt

Pest controller Jari Bremer: “With image-recognition in the cage we avoid unwanted by-catches and we are more efficient, we only catch animals that pose a risk. It’s great that we’re able to catch only the right animals, thanks to a camera and technique that keeps improving.” Bremer received a signal from the cage in Winterswijk on his smartphone and confirmed the catch in the field. He delivered ‘smart life traps’ in German project areas as part of the Life Mica project. This will happen in Belgium as well.

Risk to dikes and nature

Muskrat and coypu are alien species in the Netherlands. They have no natural enemies, they weaken banks and dikes and they interfere with inland species. In the past half century the muskrat population declined from half a million to less than 10%. In 2021, fewer than 45,000 muskrats were caught in the Netherlands. Hundreds of coypu are caught in the Netherlands along the border, but they are hardly found further inland.

Up to the border

The water authorities want to push back the muskrat up to the border alike. “Technical innovations will help”, says Pascal van der Linden, team leader pest management. Such as image-recognition, which now yielded the first catch. “This was preceded by a period in which we created an extensive image bank with all kinds of species, modeling and data analysis. For example, the camera has also been tested in a German otter center.”

50 traps

A total of 50 traps will be rigged with image-recognition based on AI software, supplied by Robor Electronics bv in Bentelo. The ‘smart life trap’ will further develop in the coming years. Muskrattenbeheer Rivierenland works on the Dutch-German border between Haaksbergen and Groesbeek, with a working area that extends over 4 water boards between the border and the sea: Rijn and IJssel, Vallei and Veluwe, Hollandse Delta and Rivierenland.

Does DNA mapping help to manage the influx?

One of the components of the Life MICA project is a study into the interrelationship of muskrats. This allows migration routes to be determined. In the period February 2020 – February 2021, muskrat trappers in the Netherlands, Belgium and Germany collected tail tips from muskrats for this research.

Central to the research is the muskrat population in the province of Friesland. Samples were collected in Friesland and in a surrounding area. As a reference, samples were also collected in Germany (Vechte), Flanders, Rivierenland and Tiengemeten.

Research questions

Three research questions have been formulated for the study:

  1. Can source areas be recognized through a difference in DNA profile?
  2. Is there reproduction within Friesland?
  3. Is there immigration from surrounding regions?

Source areas and Friesland

The results of the DNA analysis are plotted in the graph above. Each color represents a trapping organization. Red dots between yellow dots means that the muskrats trapped in Friesland show a strong relationship with the population in Zuiderzeeland, and especially the Noordoostpolder. The 3 free red dots on the right side represent the ‘own’ Frisian population.

Origin of muskrats trapped in Friesland

3% of the muskrats sampled come from the presumably ‘own’ population of Friesland. The distribution of the origin of the other sampled muskrats is:

  • 58% Noorderzijlvest
  • 26% WDO Delta
  • 13% Zuiderzeeland

The answers to the questions:

  1. Can source areas be recognized through a difference in DNA profile?
    Yes, the DNA method used can be used to tell the different source areas apart.
  2. Is there reproduction within Friesland?
    Yes, some animals have been found from the (presumably) original muskrat population in Friesland.
  3. Is there immigration from surrounding regions?
    Yes, except from North Holland, muskrats have been found from all surrounding areas.

What’s next?

Based on the results, the trap intensity along the probable inflow locations in Friesland will be increased by placing a cordon of traps here. In addition, in the period from February 2022 – February 2023, the tail tips of all muskrats trapped in Friesland will be collected for follow-up research.

Muskrats trapped in Friesland after positive eDNA sample

In August/September 2021, 3 muskrats were caught in Friesland thanks to eDNA. It concerned an area with a low muskrat population.

Tracking down muskrats and coypu can be made easier by detecting the DNA they leave behind in the water. This DNA is called environmental DNA (eDNA) and comes from, for example, intestinal cells in faeces.


In this case, a 5 km section between the towns of Bolsward and Hartwerd turned out to be positive. Subsequently, 1 km sections were sampled, of which 1 was positive for muskrat eDNA. Afterwards point samples were taken.

Feed traces

During point sampling, the muskrat trappers found some questionable feed traces. These did not directly indicate a muskrat, but could also have come from a water vole. Thanks to the eDNA the muskrat trappers knew that muskrats had been present, so they so they carefully scanned the water’s edge. There was indeed a muskrat burrow. This is how the first muskrats of 2021 were trapped in this area.

Caught muskrats between Bolsward and Hartwerd

A. In red, positive 5 km route;
B. In red, positive 1 km localization section;
C. Localization point samples. Red: highest positive concentration, orange: moderate positive, green: negative. Arrow: location of the trapped muskrats. At the second red dot at the top, the water was too deep;
D. The trapped muskrats.

Latest progress in development Smart Life Trap: first catch

This project aims to develop smart life traps that use image-recognition. Thus we prevent unwanted bycatches of protected species as European beaver or otter and only catch target species like coypu and muskrat. Recently we made good progress on the Smart Life Trap project. The first catch was made in the field.

Image-recognition is dependent of many pictures; we received lots from the field. Thanks to people from the regional water authorities and volunteers from zoos as well.

New model

A new model was made and tested. Overall the new model already makes good predictions. The Artificial intelligence specialists made a model that is quite accurate. Below is a ‘confusion matrix’. It shows the number of accurate predictions: mainly correct ones. It predicts ‘Rat group’ when it is actually a ‘Rat group’ indeed and the same with birds and the ‘Others group’ category. However, it still made some incorrect analysis. This was improved in the last months of 2021, with new photos from the field.

Real life tests and results

Below are photos with a ‘heatmap’ overlay. The more yellow/red the image is, the more this part of the picture was used for recognition. We can see from the photos that it concentrates mainly on the actual animal even when other objects like leaves or food are in the picture as well. And recognition succeeds at night as well.

Many images

In the past months we also succeeded in entering many images of otters and beavers, with the help of a German otter centre (Otter-Zentrum Hankensbüttel) amongst others. This will make the model even more accurate, even when animals look like the target species but are different.


The camera was built into 5 traps and tested in different settings, day and night. The first actual catch in the field occurred in December: a muskrat. As a control set-up, we use 24h wild game camera’s to observe the test traps and the animals entering it.

50 Smart Life Traps

In the first months of 2022 a total of 50 Smart Life Traps will be deployed across national borders: Germany (20), Belgium (5) and the Netherlands (25).