More information
Description
Bedalov d.o.o. is a young and innovative micro enterprise and a spin off from University of Split (professors and Phd students). They have developed an extensive and comprehensive suite of machine-learning algorithms for automatic segmentation and classification of complex structures detected on the digital images. In this R&D project, Bedalov will develop and validate a custom automatic analysis service of video materials based on Artificial Intelligence (AI) and machine learning to inspect fish tanks'' fouling. They will collaborate with Njord Aqua AS, a small Norwegian company specializing in inspection of fish tanks infrastructure with their ROV (Remote-Operating-Vehicle) and camera equipment.
Summary of project results
The fish farming industry in Norway, as well as world wide, has one common challenge, namely to find the right balance between cost, pollution and fish welfare. Usually, the cleaning of the fish tanks/nets is done on the regular intervals, instead of when needed, which means that they are overcleaning the nets to be on the safe side. However, if nets are cleaned too seldom, the oxygen level in the fish tank will be reduced, which is not good for fish welfare and if they are cleaned too often, excessive marine pollution can occur. Cleaning of fish tanks/nets is usually done with high pressure equipment and when the cages are new, there are fresh coatings on them that contain copper, which is then released during the cleaning and contaminates the marine environment.
The project aims to enter the market of fish farm inspections as they have seen that there are already infrastructure acquiring image/video data, but still lots of “manual” work has to be done, where experts spend hours on image analysis (by eye), while more objective measurement would be needed.
In this small R&D project, Bedalov d.o.o. will develop and validate a custom automatic analysis service of video materials based on Artificial Intelligence (AI) and machine learning to inspect fish tanks'' fouling. They will collaborate with Njord Aqua AS, a small Norwegian company specializing in inspection of fish tanks infrastructure with their ROV (Remote-Operating-Vehicle) and camera equipment.
Throughout the project, Bedalov and Njord Aqua collaborated to make experimental time-lapse video materials for fish tanks, compared the same segments of the nets calculating overgrowth of the fouling over time, from initial timepoint with clean nets of fish tanks, until the oxygen level in the fish tank starts to drop significantly. Then they calibrated vectorized segments by training their machine-learning algorithms to determine the relationship between the fouling growth curve and the oxygen level. They have proposed optimal objective measures of growth indexes and validate them with the previous, manual approach. They have completed a service of automatic video analysis of biomaterial growth on fish tank nets.
Bedalov analyzed thousands of images with the powerful image/video recognition/analysis tool and quantified with objective measure how much fouling is present in different areas of the net. They combine the images with GPS markers and sent back the results to Njord Aqua in the form of indexes (i.e. index 4 means urgently need for cleaning, index 1-2 means no immediate need for cleaning.
Applying AI solutions to this sector would secure that net-cleaning only will be carried out on a need-to basis instead of at fixed intervals, as is the case today. This will reduce fish farming maintenance costs, improve fish-product quality and quantity, and other animals/plant welfare (secures sufficient oxygen levels), and protect the environment. The seabed below the cages has a limited capacity as a recipient. Copper is known to reduce to natural decomposition of feed spillage and faeces from the fish.
In the short term, Norway will be the main market and Njord Aqua will be the first users of the new services. As a first step after the project completion, Bedalov plans to sell service to analyse overgrowth of fouling to the inspection companies. Later, depending on frequency of image material uploaded in their database, Bedalov plans to charge monthly or annually subscription fee for using the software. Each fish tank will have a dedicated address on Bedalov''s server. Bedalov will maintain, upgrade and improve the software on their server and thereby improve the services for the customers (fish producers or inspection/maintenance companies).
Summary of bilateral results
As a bilateral outcome, the business partnership with the Norwegian partner Njord Aqua AS is expected to continue after the project implementation. Promoter was responsible for AI and machine learning software, and the Partner was responsible for hardware (fleets of ROVs remotely operated from the Bergen area).