Nutreco Feed Tech Challenge 2018

Submitted
Approved
Gathering votes
12/20
Votes
20 votes
Submit full business case
Gathering votes
Top 10 selection

Improve TGC and FCR in salmon farming.

Since the start of commercial salmon farming, one of the challenges has been to get a consistent high growth rate (TGC) without having to compromise the feed conversion rate (FCR). As an example the average TGC for the last 10 years, in the Faroe Islands, has been just above 3.0. The variation can be as much as 2.5-3.5.

 

The salmon farmers have spent a lot of money and time on equipment to get more consistent TGC. Among the equipment is underwater cameras, using people to watch live video streams. But it's a known issue, that humans can't be concentrated on a task for more than 10-20 minutes at a time. The problem gets even bigger when people have to focus on many tasks (video streams) at the same time.

 

Describe the problem your company wishes to solve and how your product or service will solve it

FaroeSea started as a camera system developer and supplier to the aquaculture. But after some time, the improvement of the biological results were not as expected. The problem was the human concentration ability and people's ability to interpret whether there's over or under feeding on a video stream. The solution was computer vision.

Approx. 50-60.000 Ton salmon have been slaughtered the past 3-4 years where SeaV Detector has been used to decide to increase or decrease the feeding rate on a daily bases. The results showing a TGC=3.2 - 3.5, meaning 1-1.5 kg extra fish per smolt compare to the average TGC.

But still it is the feeding manager that do the adjustment on the feeding system. The next step is a fully automated system. That is what SeaV SmartFeeder is, and it has been in operation for the past few month's on two salmon farms.

Describe other business assistance that you are seeking from Nutreco

Get a test trial for SeaV  Detector and SeaV SmartFeeder on salmon and other species of farmed fish.

Are you available to participate in the final event in the Netherlands 28, 29, and 30 May? All expenses will be covered by Nutreco

Yes

Describe how the prize - a validation trial in our facilities - could boost the development of your business

Next step is to develop SeaV SmartFeeder to use machine learning to make decisions on a right feeding regime. One parameter for machine learning to estimate right regime for optimal TGC, is the feed ingredients.

If you already have a website for your business, please share the URL

www.faroesea.com

 

Tagged users
edited on Apr 1, 2018 by Bartal Andreasen

teresa debesa 2 weeks ago

The idea has been progressed to the next milestone.

Reply 0

Chris van Bussel 2 weeks ago

HI Bergur, great idea! I think it is one of the most boring jobs on a salmon farm. Just watching a live stream and decide up or down regulate feeding. It is indeed one of the biggest cost factors thus a fully automated solution is extremely useful.

How do you want to apply machine learning? Do you want to watch the salmon to see their behaviour adaptation due to feeding, or do you watch un-eaten pellets at the bottom of the cage? What ideas do you have here, could you enlighten a little bit more?

Regards,

Chris

Reply 1

Bergur Andreasen 2 weeks ago

Hi Chris. Thanks for the comment.
We are using Computer Vision to analyze the video streams in real time from each cage. When a certain amount of pellets are detected per sec, the software push a notification to the Feeding Manager to lower the feeding rate or to the Feeding System to automatically lowering the feeding rate.

In the new SmartFeeder module, we will collect data, like the time of the day when pellets are detected, how big the dose was at that time and feeding rate in kg/sec. Other data like weather, current, oxygen level and the composition of the feed will also be collected. We are planning to use machine learning to analyze these data to predict the feeding in the
future both short and long term.

Reply 1

Chris van Bussel 2 weeks ago

Hi Bergur, I fully agree. Long term prediction of feeding (say on weekly or monthly basis) is not easy, but we can do it nowadays. The day by day (or hour by hour) feeding on appetite is something different. Day to day variations in appetite/feed intake cannot be explained until now, so trying to understand this is very interesting in term of feed-efficiency!

Reply 0

Bergur Andreasen 2 weeks ago

Hi Chris. In commercial farming we have to rely on people's ability to do proper feeding. People are different, and I believe that's one of the reasons why we get different biological results. With the detector system we get more hard evidence on the feeding. All in all this should give better foundation to predict the feeding
and make a better feeding regime.

Reply 0

View all replies (3)

Khalil Safaei 2 weeks ago

Very Well!

Reply 0

Khalil Safaei 1 week ago

Good luck!

Reply 0

Share