Final Report Summary - FISHSCAN (Development of novel system for continuous remote monitoring of weight, growth, and size distribution of fish in aquaculture enclosures) Executive Summary:FishScan full title is “Development of novel system for continuous remote monitoring of weight, growth, and size distribution of fish in aquaculture enclosures”. It is a 2 year project from 1. June 2011 and is funded by the EC, FP7, research for SMEs. WP1 Enhanced scientific understanding and product design specification: T1.1 End-User Specification: An end-user survey was conducted to establish a set of end-user specifications for the FishScan biomass estimator system. T1.2 Product Design Specification: The detailed specifications for the complete FishScan. T1.3 Behavioural studies of fish in relation to external factors: The conclusion of the study is that is not expected that the use of high intensity artificial lights will have any long lasting negative effects on the behaviour or welfare of Atlantic salmon or European sea bass. WP2 Design and construction of underwater and topside units: The overall technical objective of WP2 was to develop robust housings designed to operate in marine conditions. T2.1 Selection of materials for FishScan: Materials suitable for underwater and topside housings were recommended based on materials characteristics. T2.2 Design and construction of underwater housing: Two separate light units and the camera unit were designed to have flexible positioning. T2.3 Design and construction of topside unit: The selected enclosure containing the electronics was non-corrosive, impact and intrusion proof (IP66).WP3 Development of software for volume estimation: Our target was to estimate the volume of fish from the ToF depth (3D) images. A simple model (e.g. ellipses and ellipsoids) of the fish has a very crude approximation and results in large errors. We have used a more accurate 3D fish model provided by TI. Movement of the fish are mimicked using a patch-based deformation “As-Rigid-As-Possible” algorithm based on correspondence between the fish model and the depth image. An initial transformation locates the fish in the image and determines its pose using key point detection on the head, upper fin, lower fin and tail. Another algorithm computes the length and volume.WP4 Electronics and control system: Task 4.1 System specification and components: Deliverable D4.1 describes the system specification and components related to the FishScan electronics. Task 4.2 LED driver development: A fully functional LED driver was developed. Task 4.3 Control board development: A fully functional control board was developed. Task 4.4 Prototyping, debugging and test: The camera unit, LED unit and tops ide unit were integrated, debugged and tested. WP5 System integration and industrial validation: Task 5.1 PDS follow-up and risk management: Product Design Specification was updated during the system integration. Risk management was handled by continuously. Task 5.2 System integration: Integration of camera, sub sea unit and volume estimation SW. Task 5.3 Small-scale validation of complete system: Small scale validation was conducted in controlled conditions with live fish, in indoor tanks and outdoor fish cages. Task 5.4 Large-scale validation of complete system: The complete FishScan system was successfully validated in large-scale at industrial salmon farms in Norway with promising results regarding sufficient accuracy of volume estimation.WP6 Dissemination and use/exploitation activities: Task 6.1 Dissemination of knowledge: The FishScan consortium has through their extensive network played a key-role in dissemination towards the aquaculture industry and other sectors from the very start of the FishScan project. Task 6.2 Training and transfer of knowledge: Training and transfer of knowledge was given high priority throughout the project. Task 6.3 Protection of IPR: Exploitation rights have been clearly defined in the Final Plan for the Use and Dissemination of the foreground (D6.5). Project Context and Objectives:The FishScan Research for SMEs project seeks to increase the sustainability, profitability and the competitiveness for the European aquaculture industry. FishScan will develop a system for continuous and highly accurate remote monitoring of weight, growth and size distribution of fish for use within aquaculture enclosures, primarily sea cages. The aim is to achieve an accuracy of 99 % in size measurements of fish larger than 1kg and 97 % accuracy for fish at 100g. A working prototype is built, tested and industrially verified. FishScan is focused on training SME staff in its application and operation to make the new technology and knowledge available to all participants. The higher accuracy in the FishScan system compared to that of current state-of-the-art technology will potentially reduce the error in biomass estimates from the current ±5 % to ±3 %, and the economic benefits using the FishScan system is appraised to € 41 million annually in increased profit for the European aquaculture industry. The potential savings using the FishScan system represents € 40,445 p.a. per fish farm license. This corresponds to potential economical benefits from the project of € 41 M annually in increased profits for the European aquaculture industry. The SMEs in the FishScan project expects to accumulate a free cash flow of € 1.6 M from sales after the 7th year the product is on the market. This equals net present value (NPV) € 3.9 M in when the product arrive at market (year 0). The challenges in the European aquaculture industryGlobally, the aquaculture production is growing more rapidly than any other food sector. The European aquaculture sector employ around 70,000 full time jobs , and in 2007 there was produced around 1.6 million tons of fish in Europe, valued at €4,635 M . However, the percentage of seafood self-sufficiency of the EU is estimated to have declined from 53% in 1997 to 40% in 2006. The main reason is an increase in seafood consumption per capita and continuous reduction in catch quotas for capture fisheries. Furthermore, the aquaculture industry in the EU has not been able to increase its production to maintain the overall fish production. This has lead to an increase in fish imports to meet with the seafood demand in EU. The value of EU aquaculture exports was 10 % of the imported seafood in 2007. The growth in imports of low-price freshwater farmed fish from Southeast Asian countries has increased from 2,100 tons in 1999 to over 393,000 tons in 2007. This is mainly a drastically increased imports of Pangasius from Vietnam, and as an example 126,000 tons were imported in 2006. The EC launched a new aquaculture sector strategy in 2002 that aimed at expanding the sector in Europe. The primary aims were to increase the economically self-sufficiency of the European aquaculture industry, and to secure employment in fisheries dependent areas by creating 8,000-10,000 new full time jobs between 2003 and 2008. To reach these goals, it was estimated that an annual 4 % growth in community aquaculture production was required. However, by 2009, this has not been materialised at all. The European aquaculture industry need to grow top meet the European and global seafood demand. This growth must be sustainable in the long run. It is of vital importance to maximise the utilisation of feed, to optimise fish growth and production yield, to be able to plan the production to meet optimum market requirements and best financial performance. To achieve this there is a need for improved systems for biomass estimation.The need for biomass control in the European aquaculture sector.Fish farming have not always been seen as an equal player with other food industries including terrestrial animal farming. However, cultured fish are domesticated animals on the same level as farmed terrestrial animals. Thus the farmer must have control on the number of fish in stock and their growth and health. There is no other form of animal food production where such a high biomass is concentrated in such a small area as in fish farming. This represents a huge challenge for the fish farmer in monitoring his stock as the possibility for visual inspection is limited. To have control of the farmed biomass, the number of fish in stock must be counted, and a representative sample of the fish population must be weighedI. In modern sea cage culture, it is a resource demanding challenge to manually catch and weigh a representative sample of population consisting of several hundred thousand individuals. One of the main challenges for the European fish farming is the increasing price of fish feed. The majority of fish cultured in Europe are carnivorous species. Feed for carnivorous fish is mainly based fishmeal and fish oil (FMFO). The sustainability in use of FMFO is debated. It is therefore important to optimise feed utilisation in fish farming to avoid waste of valuable natural resources. Feed constitutes 50% of the production costs in fish farming (i.e. feed cost per kg produced fish is €1.07 in salmon and €1.84 in sea bream cage culture). The cost of excess feeding is therefore very high. Additionally, uneaten feed may lead to organic build-up under the sea cage resulting in anoxic conditions, or it may end up as dissolved nutrients stimulating algal growth that is potentially harmful for the environment and the for farmed fish. If the fish is given too little feed, the production capacity is not reached ,and some fish always gets what they need at the expense of others increasing the size distribution of the fish. This leads to loss for farmer and the fish processer due to high number of fish in suboptimal size. Too little feed also result in higher aggression between farmed individuals (fighting, injuries) and violation of the general animal welfare acts. The sustainable solution is exact feeding, which can be achieved by the following two strategies.• Feeding based on appetite – requires adequate control possibilities such as visual inspection and/or automatic detection of uneaten feed pellets.• Feeding based on tables (made by feed manufacturers) – requires accurate biomass control.Which strategy to follow depends on several factors like fish species, fish size, season, and variability in environmental conditions. Very often a combination of the two strategies is recommended, thereby making modern fish farmers dependent on exact bio mass control knowing the number of fish in every single unit, their mean weight, and the size distribution. The fish producerIn most countries, a licence from governmental authorities is needed to farm fish. Such licences very often put restrictions to the amount of fish that can legally be produced at any site. In Norway, aquaculture permission for salmonidaes include a production limit at 780 tonnes per licence (a fish farming company may have several licences). The standing stock cannot exceed this biomass and farmers profit may depend on the ability to balance close to the upper limit without breaking it. The standing biomass has to be reported to the authorities on a regular basis. Thus the need of improved biomass control is related both to farmers’ economical performance and to the governance of the aquaculture sector. A study performed in 2009 showed that on average there is a ±5.3 % error in biomass estimations of farmed salmon and rainbow trout (the error was up to ±40 % in some cages). The economical losses related to wrong estimations will differ between an underestimation and an overestimation: • 5% overestimation of biomass on one license (production capability of 1200-1300 tons) cost the farmer € 140,650 in excess feed, and another € 16,250 in loss of sales income, adding up to € 156,900 p.a. • 5% underestimation of biomass on one license (production capability of 1200-1300 tons) cost the farmer € 6,500 in extra in sea lice treatments (to low sea lice treatment dosage may require an additional treatment), and another € 16,250 in loss of sales income, adding up to € 22,750 p.a. The reason for the large difference between under and overestimation is that it is easy to calculate the cost of excesses feeding, but difficult to estimate the cost of unused production potential as a result of underestimation (i.e. the loss due to unused growth potential in a fish stock that receives too little feed for optimal growth). An experienced fish farmer may notice that the fish are not fed enough and appetite feed the fish to be able to utilise some of this potential. However, fish farmers focus on keeping the feed conversion rate (FCR) low and minimising excess feeding. This makes it is difficult for the farmer to deviate too much from the feeding plan (feeding based on tables). Furthermore, it will not be possible to utilise the whole growth potential of the fish as a farm license have an upper production limit which may be exceeded by adjusting up the feed plan. The farmers would be able to utilise more the potential for growth in the farmed stock if they had better control with the biomass as they would be able to trust the production plan and feed table. The average loss from ±5 % error in estimation of biomass is € 89,825, which leads to an annual € 91 M loss of profits in the European sea bream, sea bass and salmonidae cage farming industry.The fish processerThe fish processer start to plan how to organise the production line one week before harvest based on estimations of biomass and size distribution provided by the farmer. These estimations are thereafter updated daily until harvest. However, estimations will deviate from the real numbers. The greatest challenge for the processer is to deal with a batch of fish that differ considerably in size (wide size distribution), as this may lead to suboptimal utilisation of the processing capacity. This represents a loss to the processer as capacity will not be utilised efficiently. The sales linkIndustrial aquaculture has resulted in a wider use of long term forward contracts including details about both volumes and sizes. To be able to fulfil the contract requirements the farmers can tolerate only minor deviations from what they believe is delivered to the fish processer and what is actually received. Large deviations may result in costumers not getting the agreed commodity and or a big volume of slaughtered fish without any customer at hand. Around 80 % of all sales of farmed salmon are arranged pre-harvest and dependent on correct estimations of average weight, size distribution and total volume. Errors in these estimations will lead to loss as the optimal price will not be obtained as the extra fish may be sold on the spot market to an unknown price. If the real biomass is less than estimated or the size distribution is not as agreed, the customer may insist that the contracted amount of fish and size distribution is obtained. Thus the farmer may need to dump the unwanted fish on the spot market at a low price, before getting hold of the fish need to fulfil the contract. This is expensive and reduces the profit by €0.13 – €0.50 per kg in the sales link. How to obtain biomass control?Fish are typically counted at several stages in the production chain; Hatechery – Smolt Production – Transport – Ongrowing – Processing - Market. Counting of juvenile fish is performed in combination with vaccination and size grading in land based hatcheries and smolt production plants. The fish is also counted in relation to transport in live fish carriers from the hatchery or smolt plant to the sea cage. Current available counting technology has high accuracy (e.g. www.aquascan.com). Such automatic counters are often connected to a fish pump and used when transferring fish from one culture unit to another. The fish may also be counted as it is allocated from one cage and dived into several cages (either for size grading proposes or for thinning the fish density). The system is very accurate as long as the operators have good practices on regular calibration of the system. , , The fish farmers is also expected and imposed by the authorities to have good practices regarding removing, counting and keep strict records of number of dead fish. Reliable equipment for removing dead fish (mort extraction systems) is also available to a reasonable price (e.g. www.liftup.no). Thus it is possible to accurately count the number of fish at several stages through the sea cage on growing period. Typically around 250,000 smolts are stocked in one sea cage at approximately 100g size (fish density 2 kg/m3). When the density reach 18-20 kg/m3, the fish is counted and split into several cages for further on-growth, where the density typically start at 6 kg/m3. The fish may stay in this cage until harvest size (4-5 kg) is reached after an on growing period of about 72 weeks (density may reach 25kg/m3). However, if the size distribution of the fish is suboptimal (i.e. with a high portion of the fish distributed widely outside the optimal harvest size), the farmer would profit by size grading and counting the fish to allow the fish at optimal size to be made ready for harvest and to allow the smaller fish to grow to optimal size. FishScan – the solutionThe FishScan system monitors weight, size and size distribution and gives continuously figures with high accuracy with little need for human resources. Estimated cost of a system is €18,000 which makes it economical for the farmer to have one FishScan unit permanently installed in every single sea cage. The system will be made compatible with existing farm control tools for production monitoring. The working FishScan prototype comprise of a camera module, a system for vertical and horizontal movement of camera unit within the sea cage, a top side computer with software for image processing and volume estimation that is communicated to the farm control tools. . To make the new technology and knowledge available to all participants, FishScan is focused on training SME staff in its application and operation.Enhanced competitive positionThe proposed FishScan project shall have higher accuracy than competing systems currently on the markets. The aquaculture industry is currently asking for such technology, and will improve the competitive position of both the participating SMEs, as well as their aquaculture costumers. This will enable better competiveness compared with Asian aquaculture companies in particular, which in general have access to low cost labour resources in a much higher degree compared to European companies. European fish farmers have been facing serious competition from low cost countries during the last 2-3 years . This especially applies to the white fish sector (sea bass, and sea bream) in southern Europe. A competitive European aquaculture industry is a prerequisite for an existing and competitive European aquaculture equipment industry. Through the development and implementation of the FishScan system, the European aquaculture industry will facilitate both. Several Scientific Objectives and Technological Objectives (S&T) have been identified to be able to develop the FishScan system. These S&T Objectives are presented in the following table, which also identifies how the objectives are linked to Work Packages’ (WP), Deliverables (D) and Results. Scientific objectives are: • Identify end-user requirements through an end-user survey among fish farmers and equipment suppliers (WP1 - D1.1 Result 1). • Identify the performance requirements of the technology and systems to be developed in the project, resulting in a Product Design Specification (WP1 - D1.2 Result 1).• Identify special issues and challenges necessary to account for when transferring Time-of-Flight technology and software to an aquaculture setting, including illumination.• Increase knowledge of farmed fish behaviour in net cages, focusing specifically on how the behaviour is affected by foreign objects, artificial light, and by varying environmental conditions, to be able to prepare optimal procedures for representative sampling of fish in net cages of varying sizes (WP1 - D1.2 Result 1).• Develop a conversion model for calculation of mass from volume. Investigate available data on this relationship for the two target species, Atlantic Salmon and Sea Bass. The model will either be in the shape of a mathematical formula or as a look-up table and will be implemented in the control system to be developed in WP4 (WP1 - D1.3 Result 1).Technological objectives are: • Select materials and components for the FishScan system (WP2 - D2.1 Result 2).• Design and construction of underwater housing for ToF camera, a video camera, LED light sources and LED driver electronics which complies with the technological requirements defined in Task 1.2 Product Design Specification (WP2 - D2.1 Result 2).• Design and construction of topside unit for proper housing of image-processing unit, control system and power distribution unit (WP2 - D2.1 Result 2).• Design of robust camera module to optimise LED light source power and positioning, while minimising unwanted scattered light, and additionally maximising reliability and lifetime for application in marine environment (WP2 - D2.1 Result 2).• Development of software, based on existing 3D camera software, which will be optimised for use on fish, particularly targeting issues such as blurring due to movement, variations in colouration of individual fishes and other occurrences for which the original software is not designed to handle (WP 3 - D3.1 Result 3).• Development of LED driver electronics (WP4 - D4.1 Result 4).• Development of a control board for camera and user communication interface, processing unit, non-volatile memory for storage of fish data, external sensor interface and power supply distribution (WP4 - D4.2 Result 4).• Integrate and validate the FishScan system to obtain a working prototype (WP5 – D5.1 Result 5).Project Results:WP1 - Enhanced scientific understanding and product design specificationTask 1.1 – End-user survey and specification:An end-user survey was conducted to establish a set of requirements for the FishScan biomass estimator system being developed. Teknologisk Institutt was in charge of this activity. In-person and phone interviews were held with selected industry actors about fish biomass estimation, resulting in development of an end-user product specification. Professional key aquaculture representatives participated with their extensive knowledge. A set of definitions was made in order to make sure all participants had the common understanding of key issues in the survey.Focus of the end-user survey and specification was on issues covering accuracy of estimation, capacity, degree of automation, power supply, ease of use, cost, durability, environmental and user safety, and compatibility with established production planning routines. A rating regime was implemented to provide a precise and measureable differentiation of opinions. Data/information was extracted and organised from survey results and evaluated with basis on opinion variations and commonalities. Results were concluded in form of an end-user product specification (Deliverable D1.1).A selection of end-user requirements:• The system should have a good price/performance ratio.• The system must be user-friendly.• The system must be small and light enough to be carried and installed by one person.• The system should easy to maintain by operators.• Housings must be easy to clean and disinfect.• Frequency of regular service requirements should be low.• Service time should be low.• A handheld unit for displaying biomass data on site should be available.• The system must perform in a variety of underwater visibility and current conditions.• Working range of the subsea unit must be minimum 0-10 meters underwater, but possibly be dimensioned up to 50 meters.• The system must be safe to operate.The end-user product specification was the basis for developing the Product Design Specification (Deliverable 1.2). Task 1.2 – Product Design Specification (PDS):Based on the end-user product specification established in Task 1.1 a complete technical product design specification has been developed. The specification of the complete FishScan system was broken down to subsystems completely describing each of the work packages. In order to ensure that the development work to be performed in the project is feasible, the end-user product specification has been evaluated with basis in past experience, theoretical limitations and cost-benefit considerations. The PDS includes requirements regarding functionality, performance, size, operation, maintenance, reliability, safety, cost, environment, appearance, documentation, and compliance to relevant standards.A selection of requirements: • The complete FishScan system should be patentable.• Methods and sub-components should not interfere with existing patents which block free use in FishScan.• Fish stocking densities up to 25 kg/m3 for salmon trout and sea bass in sea cages, and up to 80 kg/m3 for sea bass in tanks/RAS systems.• Weight ranges for salmon 0,1-12kg, for sea bass 0,3-1kg.• The most important parameter for end-users is average weight, before size class distribution.• The system must automatically calculate fish mass from volume which comes from 3D data, and possibly 2D data. • Accuracy requirement for average weight measurements is 98%for salmon/trout and 95%for sea bass. • Accuracy requirement for weight class distribution is varying between 95% and 98% for salmon/trout.• The overall weight of the system should not exceed 30 kg’s, more specifically up to 18 kg for the topside unit and 12 kg for the subsea unit. • Subsea housing must be designed to be accepted by fish.• Housings should look modern and robust.• Housings should be easily identifiable as a FishScan/Storvik Aqua product • All materials should be environment friendly which includes use of degradable and/or recyclable materials which at the same time are durable against sun/rain/temperature/water.• The camera unit must provide good quality images, meaning focus on whole fish and no occultation between fish or other objects. • The electronics control board must provide interfacing with cameras, input from sensors and communication with topside unit.• The system should provide information about sea temperature, camera tilt, compass direction and pressure/depth.• Only one (combined) cable should be needed for data and power transfer between subsea and topside units.• Parameters that should be provided by the system to the end-users are average weight, weight class, condition factor, growth.• The lighting must not be harmful for the fish during normal operation over extended time.• FishScan electronics unit must provide wireless data transfer (e.g. WLAN, Bluetooth, GSM).Task 1.3 – Behavioural studies of fish in relation to external factors: Potentially dangerous stimuli, for example a sudden sound, movement or bright light often elicit a startle response in fish, which is typified by increased movement away from the stimulus. If the stimulus persists without negative effects then animals can learn that it is not dangerous and stop responding (they habituate). This study focused on providing essential knowledge on whether fish behaviour is affected by artificial light of different intensity and patterns, in order to inform the design of the FishScan system and the final choice of light and camera technology. The main questions are the following: • How do fish respond to the light - by attraction or avoidance?• Do responses vary with depending on pattern and/or intensity of light?• Does the light compromise welfare of the fish?Atlantic salmon and European sea bass, of two sizes, were exposed to increasing intensities of green light projected into a tank through a side window. Different light patterns and intensities were tested, full image and striped image. The behavioural responses of the fish to were recorded remotely and analysed manually using video footage to allow quantification of the distribution of the fish within a tank. Avoidance was determined by assessing percentage of fish in the cone of light before and after the image was projected. Changes in distribution can represent how the fish respond to the light source and how this changes with time. Results indicate that fish that are exposed to bright green light exhibit a visible startle response and avoid the light area initially. Nevertheless they habituate after ca. 2 hrs., gradually returning to the lit area. There is also an indication that groups of fish that are exposed to the light source on consecutive days have some memory of the light and habituate faster. No difference in the response to the light can be detected between fish species or fish size, indicating that behavioural responses are constant. There is a strong avoidance to the striped light pattern shown by both fish species with little or no habituation evident. It is not recommended that striped light pattern be used. Conclusion is, that is not expected that the use of high intensity artificial lights used in the camera technology will have any long lasting negative effects on the behaviour or welfare of Atlantic salmon or European sea bass.WP 2 - Design and construction of underwater and topside unitsTask 2.1 Selection of materials for FishScan:This task was about selecting materials for the FishScan units. It concerns the topside unit containing the processing hardware for the camera, the battery/power connection and other connection possibilities. It also includes the underwater housing known as the camera module. The aim for this task is to find proper materials for these units.The requirements for these materials are:• All materials must be non-toxic and must not release any substances harmful to either the consumers or the farmed animals (e.g. from corrosion etc.).• All materials must withstand constant sea water.• All materials must be UV resistant.• All materials must withstand temperatures between -30°C and 50°C.• Plastic materials must be suitable for injection moulding or other relevant production method.• Materials shall be light weight.• Materials shall have dark colour.Further other points were added as desired:• Materials intended to come into contact with fish must be ‘food grade’ approved• Materials in contact with water, feed and fishes shall have smooth surfacesOn the basis of these criteria polymer materials were evaluated as the key material for both topside and underwater unit. The largest challenge with polymers is the environmental degradation, which is divided in chemical degradation (oxidation and hydrolysis), physical degradation (thermal, photo and mechanical) and biodegradation. The material investigation focused on general and advanced engineering polymers. After research into polymers the investigation went into steel, stainless steel, aluminium and titanium for optional use and other components.Conclusion and Achievements:The investigation into polymer materials ended in recommendation of several possible types.The main components polymers recommended are TIVAR 1000 (high polymer weight polyethylene) or KETRON (PEEK), which both are used for marine applications. Further polycarbonate can be used for translucent components as the camera lens window(s).Concerning other components stainless steel grades as S316 or aluminium is recommended. Task 2.2 Design and construction of underwater housing:This task was to develop a design for the FishScan underwater module. The design will conform to the requirements in the PDS defined in Task 1.2 and will contain the ToF camera, a video camera, LED light sources and LED driver electronics.The underwater housing was designed to meet the most important requirements in the PDS defined in Task 1.2. Main requirements for the underwater housing were related to functionality, flexibility, waterproofness, robustness, cooling of components, cleanability, connectivity with other units, cost and ease of manufacture. In addition to the housings, the underwater unit is comprised by a ToF 3D-camera, a greyscale 2D-camera, LED light sources and LED driver electronics. Several concepts for housings were developed by TI, which were presented to the Management Board and evaluated. It was necessary to know the exact dimensions and heat-related properties of the electronics components that were going to be mounted into the housings to make the final design. Based on this process, a final concept was selected and manufactured to be used for the FishScan prototype. A solution using two separate light units was chosen to easily achieve a high degree of flexibility with regards to positioning the units in relating to each other (distance and angle). This also allowed reduction of weight and ease of transport/handling compared to a single unit with such adjustments possibilities. The three units were positioned and fixed as desired on an aluminium rod measuring approx. 2 meters. The camera and electronics inner support frame was drawn in 3D CAD software and constructed using a rapid prototype 3D printer in PVC material. This job was done at TI’s facilities in Oslo. The outer housings for camera and lights were made by a mechanical workshop according to the specifications provided by TI. A massive aluminium bolt was used as basis for manufacturing the different components. The underwater housings were constructed to be waterproof. They are dependent on the use of rubber seals (O-rings), as commonly used for subsea applications. The front cover is made of acrylic glass which protects the inside components from impacts and water, as well as providing proper transparency for LED lights and cameras. Waterproofness tests were done both at TI’s facilities as well as at Storvik before starting the industrial validation. Identified leakages were fixed by adding further sealant as well as filling the three units with nitrogen gas after vacuuming. The underwater proved to be functional, waterproof as well as providing sufficient cooling throughout the validation conducted in WP5. More details about the development of FishScan underwater housing is available in D2.1 chapter 3.Task 2.3 Design and construction of topside unit:This task was to design and construct of housing for the topside unit. The design will conform to the requirements in the PDS defined in Task 1.2 and will contain the computer for image processing, connectors for data signals and power.TI presented several alternative concepts for the topside unit. Different sizes, heat-transfer solutions, materials and mounting options were considered. The main functionality of the topside unit is to provide suitable housing for the FishScan image-processing unit and power distribution unit. It provides connectivity with external systems, such as grid power and data/power cabling to and from the underwater unit. The selected enclosure (ABB Gemini) was modified to provide connectivity for an external long range Wi-Fi antenna, which was supplied along with the topside unit. The enclosure contains delicate electronic components which in this case operate within limited in a challenging environment regarding temperature, humidity and salinity. The enclosure is impact proof according with an Ingress Protection rating of IP66. This implies total protection against dust, as well as being waterproof. A strong water jet, rain or sea spray directed at the enclosure from any direction will not have any harmful effects. This applies as well to all cable connectors, lamps and switches used. The enclosure is 100% non-corrosive, which is crucial for this application. More details about the development of FishScan topside unit are available in D2.1 chapter 4.WP 3 - Development of software for volume estimation The primary objective of this work package is to find the volume of the fish using the acquired Time-of-Flight (ToF) depth images. Task 3.1 - Direct Fish Reconstruction:The initial plan is to fit ellipse and ellipsoids into the fish from the depth image. However, this generates a huge error in accuracy. We replace this task with filtering and segmentation methods that are used as a pre-processing process in Task 3.2 and 3.3. Intuitively, our goal is to segment the closest fish from the camera and continuously segment the succeeding closest fishes. It all begins with a filtering procedure that tries to smoothen the noise in the image by using a Gaussian filter incorporated with the ToF amplitude image. In addition, we also do background thresholding where all pixels greater than 2 meters are removed, and background detection where all pixels with large standard deviations with respect to its neighbouring pixels that signifies noise are also removed. Then, we take the closest pixel to the camera and begin our region growing algorithm to segment the closest fish. The results from these steps are a number of pixels of a single fish on the image.Failures in segmentation often happens when noise is abundant that may be caused by the reflective property of the fish’s scales that creates an aura effect; by the poor visibility in water; or, by the location of the fish from the camera where objects farther from the camera produce more noise. Thus, from these steps, it is important to be able to mathematically evaluate the results and determine if the segmentation is good or bad. With regards to the amount of noise, we found out that reflectivity of the fish that induces noise have high values on the amplitude image. As a result, these images are discarded due to the extreme amount of noise on the image. Furthermore, we noticed that the amount of noise increases as the fish goes farther from the camera; so, the average depth of the fish must be less than 1.5 meters. Note that the noise is also removed by the background detection in the filtering step. Lastly, to identify whether the segmented fish has a sufficient amount of information, we make sure that the segmented fish has greater than 1,500 pixels. Therefore, all segmentation results that do not satisfy these criteria are not considered for the next steps.Task 3.2 - Deformable Fish Model:We have a 3D fish model provided by TI and, in the first reporting period, we developed a patch-based deformation model which is based on fitting the surface of the model into several patches and rigidly deform the patches to move the fish. This is a powerful and robust tool but it is a time consuming process. Thus, this requires us to develop another deformation model that is based on “As-Rigid-As-Possible” algorithm which is not only simpler and much faster but also powerful enough for our application. Its deformation is based on the point correspondence between the fish model and the 3D points from the segmented fish where the resulting deformed model have the same shape as the 3D points. Hence, the volume of the fish on the image is the same as the volume of the model. Furthermore, we also use the fish model to create synthetic depth images using Blender for our machine learning process in Task 3.3. It has the advantage of control with regards to the number of images, different location of the camera with respect to the fish (viewpoint) and different sizes of the fishes. It also has the advantage of getting accurate ground truth parameters from the image such as key point locations, volume and length. Contrary to using real depth images produced by ToF or Kinect camera, the parameters like 3D key point locations cannot be accurately found and the variety of images is difficult to produce because we cannot put markers on the fish and we cannot control the movement of the fish.For the synthetic data, we fit bones (head, spine, tail) into the fish model to create fish deformation or movements, and we turn the camera around the model to view it from different angles. It is important to note that the synthetic depth image from Blender requires us to set camera parameters which are the same as the Kinect or ToF camera to generate realistic images. These parameters include the focal length and the viewing angle. Therefore, the camera needs to be calibrated in water using the standard checkerboard pattern. Moreover, we use the calibration toolbox that manually locates this pattern since black and white in the image are shown as grey and dirty-white in water, and the automatic calibration often fails to find the pattern. For the ToF camera, we need to additionally calibrate the offset which is a constant value that is subtracted from the depth image to generate the correct value. We accomplish this by setting a board 1 meter from the camera and subtract 1 meter from the board as seen on the depth image. This offset is very important because this is the only basis of our depth perception where we comprehend that, when small fishes are close, we see them as large fishes; or, when large fishes are far, we see them as small fishes. Thus, if this is not set correctly, it will be hard to determine if the fish is small because the fish itself is small or it is far from the camera. Moreover, the depth is used in the background thresholding from Task 3.1. and large depth values are discarded which results in no data to segment.Task 3.3 - Robust Model Fitting:Prior to model fitting, we need to detect the fish and to determine its pose. Therefore, after segmenting the closest fish in Task 3.1 we need to identify the initial relation of the fish from the image and the model because the “As-Rigid-As-Possible” deformation in Task 3.2 only refines the deformation to make it more accurate (e.g. making the fish fatter) and assumes that the 3D points from the image and the vertices from the model are close to each other. Due to this, we use a machine learning method called random forest to detect 3D key points located on the head, upper fin, lower fin and tail. These four points are then used to determine if the fish is swimming to the left or right of the image, or if it is facing the camera or away from the camera. Mathematically, these are used to compute a linear transform of the model that prepares it for deformable fitting as explained in Task 3.2. Random forest learns the relation of the input parameters which are depth values of the segmented fish, and the output parameters which are four 3D key points; so that, when only the input parameters are given (e.g. depth images from ToF or Kinect camera), the forest can predict all the output parameters. This explains the importance of synthetic dataset in Task 3.2 because, in learning, we need both the input and output parameters, and the dataset for learning uses thousands of images that are labelled with ground truth information with varying viewpoints and varying fish sizes. We evaluate our forest with synthetic dataset and they generate accurate results especially when the fish is in fronto-parallel or when most of the fish structure is visible from the camera. In addition, we also tested the forest using Kinect camera on dead fish in air and they produce good results and was able to continue with model fitting with “As-Rigid-As-Possible” deformation algorithm until volume estimation. From ToF images, our forest generates good results as seen in 2D images but we are having problems validating the accuracy of the results because of the amount of noise. By visually inspecting the images in 3D, the large amount of noise creates an ambiguous location of the 3D key points. Moreover, the ambiguity from noise also affects the accuracy the deformation algorithm and, as a result, the volume estimation. Considering the amount of noise, we tried to remove the error propagated from key point detection to model fitting and model fitting to volume estimation by using a direct length and volume estimation. By replacing the output parameter as length and volume, we use the random forest to directly find the length and volume of the fish. We evaluate the direct estimation using synthetic dataset with increasing amount of noise and generate good results. In addition, we found out that the length is more accurate than the volume. This is explained by the fact that there is a cubic relation of the volume and length which means that there are more errors associated to the volume since it requires length, height and width. From ToF images in air and in water from a single fish, our results from the direct estimation are consistent for both length and volume. The consistency of the resulting values is important because it shows that the prediction of the parameter is precise and the conversion to mass can be factored accordingly. This consistency depends on the data acquisition setup (i.e. alternating lights or both light, upper light or lower light) and environmental factors that produces additional noise such as the clarity of the water or the reflectance of the fish’s scales which is explained in Task 3.1. As for other species, since this method learns from the synthetic dataset, only the fish model in Task 3.2 needs to be changed and the same process follows. WP 4 - Electronics and control systemTask 4.1 - System specification and components:The consortium is aiming at a ToF based solution supported by data from a 2D camera. This technology alternative seems to be superior to other alternatives for the FishScan requirements. TI has been working closely with Fibula in the specification work. And together we have visited PMD Technologies in Siegen. The deliverable D4.1 Electronics design document has been submitted. It has been decided to base the topside unit on an industrial PC or similar. The Subsea unit will consist of one main board possibly based on a FPGA, and separate modules for ToF Camera, 2D Camera and LED luminance source. For testing and optimization of the complete solution, it is found to be an advantage if the physical positions of the light source and cameras can be varied in regard to each other and the possible position of the target. TI arranged rental of a fish tank at UMB (Norwegian University of Life Sciences) at Aas, Norway, for capturing images of live fish. The purpose of this was to evaluate different 2D cameras as well as 3D ToF camera with different light sources. The tank had a diameter of 3 meter and a water column up to approx. 2 meter. 8-12 Salmon of approx. 3 kg were in the tank. Turbidity conditions in the fish tank varied naturally due to natural bio-particles from the biofilter in the recirculation system. It was also possible to add particles to the water by using dissolved clay in order to simulate “worst case” conditions of the natural (varying) environment in industrial aquaculture enclosures. TI also made a rig for calibration of 3D cameras for use in tank with water. This rig consisted of two circular discs of different diameters which were assembled surface to surface. The disc with smallest diameter was mounted to appear closest to the camera for calibration. The Discs were mounted to a steel frame with a fixed, known distance from the camera. This facilitated for calibration of the 3D cameras, and was a suitable tool to confirm that the distance and area/volume information could be extracted accurately from the raw data provided by the cameras. Footage was captured by mounting the cameras outside a window made of acrylic glass which was integrated in the tank, especially made for this purpose. Two different 3D cameras was used, both from manufacturer PMD Technologies, including their most recently developed ToF camera, CamCube 3.0. Among the 2D cameras tested was an industrial greyscale camera (Manta G-504B, 5 megapixels), as well as consumer colour cameras (Nikon D200 and Olympus E-P1) with suitable lenses. The lenses were selected to provide a field of view of 1 meter width at 1 meter distance. This was chosen to make sure the system can capture and focus on a whole fish in 1 frame/picture, as defined in the FishScan product design specification. Several hours of footage was recorded in the period January to March 2012 with assistance from SME partner Fibula. This (2D and 3D videos and still pictures) was stored on a hard drive and sent to TUM in Germany as input to their development of software for automatic volume estimation. The quality and usability of the recorded footage for FishScan will be evaluated by TUM, based on how easily they are able to extract the data needed to perform the volume estimation. The fish tank setup is available when needed for FishScan throughout the whole project period. Task 4.2 LED driver development:This task was to Design a LED driver board containing driver electronics and LEDs able to operate with modulation frequencies up to 40 MHz. The PMD ToF technology and LED driver alternatives were investigated. The PMD ToF cameras are based on analog signal mixing in each sensor pixel. The modulated light intensity is then by PMD recommended to be sinus shaped. There are two limiting parameters to consider when LEDs are modulated at high frequencies. One is due to carrier diffusion and recombination. This is a physical property in the pn-junction of the LED. The other parameter is the electrical behaviour of the LED. An LRC equivalent can be used to describe the electrical behaviour. For high frequency modulation the inductance of the leads and bonding wires and PCB tracks are be dominant. The first board designed by TI was using a dedicated LED driver circuit but did not achieve the required switching bandwidth and a complete redesign was required. The second LED driver board design was constructed based on input from the existing board from PMD using discrete components (transistors) and high speed comparators and LVDS receiver. Unfortunately there was not possible to get any design support from PMD, and some of the components that were chosen were not the same as used in the PMD LED light. The LED was able to switch up to a few MHz, limited by the high gate capacity load of the driving transistors.The third and final design was modified with dual P-N gate transistors with low gate capacitance (cgs) and simulated using SPICE. Most of the components have SPICE models except the LED’s that had to be estimated. This proved to be god enough for the simulation and matched the measurements of the implementation except for the PCB tracks inductance and capacitance that gave some deviation in the rise and fall time of the LED. Several LED board were assembled and tested. There were still some issues regarding rice and fall time deviations due to high capacity of the LED. This was solved by reducing the driver transistor resistance simply by piggy backing the driver transistors and reducing the resonance value by two. The remaining delay in switching time was related to the track width and length. The final LED drivers are capable of switching at 20MHz (approximately 25ns light pulse width).The LED track width is related to the current requirements of the LED and the size of the LED. It’s possible to redesign the driver using a separate driver for each LED, thus increasing the component count and placing the driver beneath each diode. The LED board has a cooling area at the back side of the board (soldering side). Due to the low duty cycle of the LED there is only a low temperature rise of the board. The LED driver boards were developed and tested by TI, in cooperation with Fibula.Task 4.3 Control board development:The control board in the subsea unit was initially designed to include image capturing and pre processing of data before sending the data to the top side. This added complexity to both the top side unit as well as the sub sea unit that had to be constructed using a Windows computer system with interface to the top side software. A better approach was to design a controller board that used commands from the top side unit to control the LED light and image sequencing. The LED light has two functions, to provide light to the high resolution camera and phase pulsed light to the ToF camera. The controller board was implemented with a function to synchronise to the ToF camera and to free run and trigger the ToF camera or to provide a stable light and trigger the 2D camera. When no pictures were taken the controller switches the LED to an idle position giving the same light intensity as the ToF image.The controller board consist of a power main board with a piggy back pre manufactured CPU board with an Atmel CPU. The CPU was programmed using BASCOM to receive serial commands from the top side and to operate LED, camera and report sensor status as temperature and pressure (depth)The main power consists of several parts, first the line power over Ethernet converting from +48V to +12V. The +12V was used for the LED driver and the controller board. The Ethernet power was designed using COTS switches that were modified with a common mode feed of each pair with +48V and power return using all eight wires. This is the same method used for the PoE standard (power over Ethernet IEEE 802.3at-2009) but in this case was able to provide twice the power (50W) in peek power. The switch also allows multiple Ethernet devices to be connected to the top side unit. This solution is not possible to industrialize, but is adequate for the proof of concept verification. The controller board was installed in the camera housing together with the ToF camera and the 2D camera. A small fan was installed together with the controller board due to a local heat spot around the ToF camera that required a little forced air ventilation. The camera housing is made of aluminium emerged in sea water and is an efficient cooling element provided that the air is circulated in the internal housing.The controller CPU and the ToF camera were logical attached to the top side unit using an USB to Ethernet bridge. This is a COTS device that was installed in the camera unit together with the modified Ethernet power switch. (The 2D camera uses Ethernet for control and image uploading). The controller board was tested by opening a serial terminal on the top side computer and writing command letters to the controller. The control boards were developed and tested by TI, in cooperation with Fibula.Task 4.4 Prototyping debugging and test:This task was to Manufacture PCB’s for the developed electronics units and make them operate.The camera unit, LED unit and tops ide unit was integrated at the lab by testing the interfaces and SW commands taking 2D and 3D images. Before integration each module was tested separately to verify correct function. The LED modules were tested using a pulse width signal generator and an oscilloscope with electrical and optical probes. Waveform and signal shapes were monitored and tested. The controller module was tested by sending command in a terminal window and the camera images were captured from the vendor SW on a windows computer.A software API was implemented into the image calculation software to control and capture images. This is a mix of both communicating with the camera vendor SW and direct communication with the underwater unit.Integration of electronics components and corresponding software included in this task were:Network and USB communication, installing drivers and API, Image capture, 2D and 3D, LED light control, select correct lamp and function, Retrieve images from 2D and ToF camera, Verify image quality and image data.WP 5 - System integration and industrial validationThe overall objective of this work package was integration of sub-parts and software developed in WPs 2-4 in order to obtain fully functional FishScan system. Validation of the FishScan system by demonstration of the fully integrated technology against the project objective including benchmarking compared to existing technologies and economic analysis of cost effectiveness.Task 5.1 PDS follow-up and risk management:The product Design Specification that was made in Task 1.2 describes the product development objectives and methods as good as possible in the early stages of the project. As the development work progresses, it is natural that the specifications are modified as a result of new breakthroughs or change of premises. A well defined Product Design Specification was made in Task 1.2 which was used thoroughly as support and reference document in WP’s 2-5. The main requirements were consistent throughout the project, and the few updates were documented as a part of D5.1.The risk management task runs through the entirety of the project and identifies and monitors risks associated with all subsystems in the project. By allowing risk management to be handled by one entity in the project, one secures an overall focus on risks and allows for thorough follow-up of identified risks throughout the project. Risk management was handled by continuously monitor risks, and to cover this topic during the monthly Skype/technical meetings and Management Board meetings. The project risk table was continuously updated and presented/discussed during the meetings, in which decisions for further action were made by the consortium.Task 5.2 System integration:Task 5.1 provided a good starting point for the integration process as it ensured that there were minimal changes needed in order to make the various subcomponents fit and function together. The minimum requirement of the prototype was that it should comply with the must-have requirements and be ready for small-scale trials and industrial validation.All hardware and software developed in WP’s 2-4 was successfully integrated into a fully functional prototype. The prototype consists of the following subsystems:• Underwater (subsea) and topside housings. These are the enclosures for the subsystems. There are two housings for the LED lights, and one for the rest of the underwater unit. Camera and lighting placement trials were performed to identify which gave optimal image quality. There is a separate enclosure for the topside unit, which is mainly comprised by a power supply, image processing PC and a long range Wi-Fi antenna. • 3D ToF Camera. This is a COTS unit made by PMD Tech, connected to the controller unit via USB. Distance calculation is done in the topside unit. Calculation accuracy is highly dependent of the LED light source.• 2D Camera. A monochrome 2D camera (Manta 145) made by AVT was selected after extensive investigation and trials by TI. It is a COTS unit connected over Ethernet. It can be software-controlled over Ethernet, but may also be hardware triggered for more reliable exposure, dependent on a signal from the controller, as it implemented in FishScan.• LED Light sources. The LED light source is a unit designed by TI (described more thoroughly in report D4.2).• Controller unit. The controller is designed by TI and consists of the following subsystems; Microcontroller (CPU), Synchronisation unit, Interface from the ToF camera, Interface to the LED sources, Temperature sensor interface, Pressure sensor interface, Positional sensor interface (prepared for, but not mounted in the prototype). The control board was made to control the two light sources independently. Based on various trials using the complete system, it was concluded that simultaneous blinking of the two light sources gave best performance. • TI and TUM have been collaborating on developing a working solution for camera and lighting triggering/synchronization. This was implemented as a part of the control board. Data flow from camera system to volume estimation software has been implemented as “.bin-files” containing packages of images from the cameras.• Integrated control software was made by TI as low level software connected to the underwater unit. Control and data collections was collected into one single software package, capable of collecting data both from 2D camera, the ToF camera, and in between control the controller.• PoE board. The power over Ethernet system consists of a power supply able to provide 3A over 48Volt to feed the subsea unit, and a system to feed the Ethernet cable with this power.• Ethernet router/switch. This is a 1 GBit Ethernet switch from Cisco, model SG100D-05. It is a COTS unit which was modified in order to be able to carry a high power level to the subsea unit.• USB Ethernet router/bridge. This is a USB to Ethernet bridge with two USB 2.0 ports that allows a remote USB connection to be controlled over an Ethernet link. The top side unit identifies the subsea unit USB device as a local USB device.• Temperature sensor. The temperature sensor measures the water temperature and connects to the control unit with an I2C interface. It is realized as a temperature IC mounted on the prototype underwater aluminium housing.• Pressure sensor. The pressure sensor measures the water pressure (depth) and connects to the control unit with an analogue interface connected to the controller’s ADC input. It is realized as a pressure component mounted to the case, allowing the seawater to come into the pressure membrane.• Positional sensor. The optional positional sensor measures the position of the Subsea unit and can be connected via RS232 interface. This functionality is not particularly important for FishScan, and was not implemented. • Cooling fan. A fan was added to reduce the temperature inside the controller system due to heat from cameras and power supply. • FishScan control interface was made as a command based program for image acquisition and camera/control board settings. Task 5.3 Small-scale validation of complete system:In order to gauge the accuracy of the FishScan biomass estimator, repetitive tests on single individuals was performed. The test included fish of various weight classes to confirm the range of the system and prove that accuracy is independent of size.TI led the activity with acquisition of fish images in controlled conditions. It was conducted in a 3 meter diameter tank filled with water using both 2D greyscale and 3D ToF camera. The purpose was to test and compare different camera systems and combinations, with regards to the ability to take photos of live fish underwater in different turbidity conditions. No. of fish varied from 2 to 25, and size varied between 1 kg to 3 kg (approximately). Machine vision equipment supplier Parameter (Sweden) physically attended during a small scale trial. They gave recommendations and also provided industrial 2D cameras for testing before deciding on which to use for FishScan. The project gained useful experience with system and component performance and functionality. In addition to this, small outdoor fish cages were used for tests on few/single individual fish with known morphology and weight. No. of fish varied from 2 to 30, and size varied between 0,8 kg to 4,5 kg (approximately). Image data was provided to TUM, including video pairs of 2D and 3D video. This activity continued during a period of a few months, parallel with continuous hardware and software improvements. Atlantic salmon was selected as main target species for industrial validation due to availability at Leroy’s sites and the immediate market potential.TI made a calibration rig for 3D cameras for use in tank with water. The main purpose of this unit was to be able to evaluate measurement accuracy and adjusting exposure parameters according to the set distance of the rig. TI also made a water resistant checkerboard consisting of black and white squares for calibration of 3D camera, based on specifications given by TUM who were going to perform the actual calibration. Performance of the system was also evaluated in-air to be able to compare with underwater performance. We evaluated the classification and regression with different camera setup where the objective is to have a consistent length and volume estimate. In this case, we solve for the mean and standard deviation of the prediction. Furthermore, the length and volume estimates depend on the ToF setup and the ToF depth offset which is a constant added to the depth image to determine the true depth of the objects in front of the camera. The objective of this evaluation is to generate consistent volume and length estimates of a single fish since the conversion of mass to volume is constant and can be changed accordingly. From these datasets, we found out that small fishes that are 1.5 meters away from the camera produces noisy acquisition and large fishes that are 2 meters away are also noisy. Task 5.4 Large-scale validation of complete system: It is important to evaluate the system performance in an industrial scale to show the applicability of FishScan in modern fish farms and under real-life conditions. The testing was performed at end-user facilities, for enclosures of different sizes and on fish at different stages in the lifecycle, from juvenile to harvest-size fish.The complete FishScan system was successfully validated in large-scale at industrial salmon farms in Norway. Results are promising with regards to achieving sufficient accuracy, but the volume estimation software system would benefit from even more training. The ToF dataset in water consist of multiple (April 2013) and single (July 2013) live fishes where the camera is on free-run. On the April dataset, the camera is on alternating light setup while, on the July dataset, the camera is on two lights in the new mechanical setup, with the two lights synchronized. Based on the results from April, the lower light generates worse results such that parts of the upper body are considered noise by the background thresholding. There are two other schools of fishes but, due to the environmental light conditions, there are too much light on the fish which creates very high amplitudes that generates an aura effect on the fish as shown in Figure 39. This creates a problem in segmentation which is crucial for the random forest. The dataset from July consist of one live fish in free-run using the new mechanical setup with two lights. In this case, we need to compute the offset that corrects the depths in the image. To find this offset, we use a board at 1 meter in front of the camera and subtract the pixels on the board by 1. Validation methods and results are described more thoroughly in D3.1 and D5.1.Fish behaviour observations were done during the industrial validation. Personal observations (G.A Romstad, Storvik / P. Vebenstad, TI) confirmed the overall conclusion obtained from the controlled behavioural trials conducted at Swansea University. The fish in industrial sea enclosures seem to be even less affected by the green, continuous light compared to results from fish tanks. The fish do not seem to avoid the light source, as fish show normal swimming behaviour even the first few minutes during exposure to the FishScan light. WP 6 - Dissemination and use/exploitation activitiesTask 6.1 Dissemination of knowledge:The FishScan consortium has through their extensive network played a key-role in dissemination towards the aquaculture industry and other sectors from the very start of the FishScan project. Activities included participation in seminars, presentations in major exhibitions and conferences, and publication of non-sensitive results. All partners were engaged in the protection of IPR, and it was stressed that dissemination of project results will only occur after the IPR, or “knowledge” created in the project has been protected by means of patents or other appropriate vehicles. A project web-page was established to share information of the project and its progress with the public. Promotion of the FishScan project to aquaculture companies were specifically performed through: networks of industrial contacts within the consortium; EU associations; several conference and seminar papers (ensuring rules regarding confidentiality are adhered to); presentation of project at exhibitions including: European Seafood Exposition, European Aquaculture Society Exhibition and AquaNor and dissemination through the FishScan project website.The project website (www.fishscanproject.com) was established to provide communication and information of non-sensitive information about the FishScan project to the wider public. It has been continuously updated throughout the project, having the following purposes; • Being the public face of the project• Generating interest in the project • Contributing to dissemination activities• Providing information to both project participants and the public• Providing contact details to project participants and mandate holdersTask 6.2 Training and transfer of knowledge:The RTDs ensured that the knowledge and results developed in the project were transferred to the SMEs participants with focus on fully identifying and promoting the commercial potential of the technologies and know how generated in the RTD activity of the project. Training and transfer of knowledge was given high priority throughout the project. In addition to the regular Management board meetings, both physical technical meetings as well as monthly Skype meetings (1 hour each) were held. This involved that RTD’s presented their latest work to the Management Board both orally and through status reports which were uploaded to Basecamp. These reports were usually between 5-15 written pages, and also included tables and illustrations. The most important documentation of the work conducted in WP’s 2-4 has been uploaded to Storvik’s GIT-server (including source code and documentation). All reports were continuously made available for the consortium on Basecamp for reading and input (draft/final versions). Task 6.3 Protection of IPR: Exploitation rights were clearly defined in the Consortium Agreement and the Final Plan for the Use and Dissemination of the foreground (D6.5). The consortium therefore did not feel the need to make any changes to the previous agreements. Full details of the agreement are given in Table B2 in D6.5. Potentially conflicting patents have been search and kept an eye for throughout the project, which resulted in an objection to a relevant patent application filed by a competing company to the FishScan consortium. The objection was accepted, and there are no known patents conflicting with the FishScan system.Potential Impact:With a handful of systems in operation it is demanding task for a product to brake through the barrier of sceptic which is dedicated to new products “coming to town”. As Norway has been in front of the industrial technology development for fish farming, Chile, Scotland and Canada has followed. All breeders has once or several times been guinea pigs. It is not enough for a product to be preferable in operational result, price, labour-saving, robust and user interface are parameters that end users will consider FishScan. Potential improvement is distributed in all aspects of the various systems; each system has different strengths and weaknesses as stated in benchmarking evaluations. Product specifications for FS took into account to achieve satisfactory solution of the known weaknesses of other systems. This assessment takes into account that these points are covered to a satisfactory result or will be given further development until commercial model 1. Consider the fact is that no system has been able to eliminate their weaknesses in the time FS project has been, and the market has expanded in volume FS higher timeliness and opportunity now than when the project started. Today average is ± 5% error estimation of biomass, but this can vary between 0% and up to ± 40%. Data provided from a larger breeder shows that underestimation (more fish in than you think) account for 65% of cases of error estimation. Consequences of error estimation under production expressed through feeding, medication and treatment, sorting / separation and utilization / compliance MTB. In addition, it has impact for slaughterhouse and vendors. Economic Impact of feeding provides a potential annual loss of 150 000 € per licence if over estimation of biomass 5% (believed to have more biomass then real). An underestimation of biomass of 5% (believed to have less biomass than real) can cause a production loss which could mean economic losses. At oral treatment against lice, for example, an underestimation of biomass (believed to have less fish than you actually have) cause the fish get too little medicated. This may in turn fish into an additional treatment which triggers a negative economic impact on fish producer, and can lead to greater risk of developing resistance against lice. Each additional treatment costs about £ 0.40 / kg fish. Suboptimal utilization or exceeded MTB (maximum allowed biomass) can economic consequences for fish producer. Both management and farming producer is concerned to keep level of MTB and ensure that the reported figures votes correct as possible with'' reality''. Management wants a greater extent for be able to verify reported numbers of fish producers, which is not easy with today's equipment. Error estimation of biomass in farms has consequences beyond the actual production in the sea. The slaughter / processing and sales force relies on so accuracy forecasts as possible in order to be able for extract maximum economic profit and utilization of capacity. While harvesting / processing steps are relatively flexible in terms to biomass for harvest as submitted, experiencing sales units are direct consequences of wrong estimation of biomass. This is due to the fact that the majority of fish sold already sold before it is slaughtered. This allows business units extra sensitive to deviations from the forecast. In an overestimation of biomass 5% and a loss in sales 0.25 € / kg fish sales area will potentially lose 12 million € per annum (on the basis of total biomass 900 000 tonnes). With a conservative estimate for loss equivalent to 0.13 € / kg would be the potential loss 6 million € annually (for a biomass 900 000 tonnes). Deviations for size distribution in relation to prognosis will potentially effect on sales price. This trick varies from situation to situation. According to interviewees, deviations from the forecast in size composition lead to loss of 0.12-0.5 € / kg fish. By comparing some of the consequences that could potentially lead to economic losses (feed cost, selling price and cost of lice treatment), we see that: • Overestimation of biomass 5% can potentially lead to a loss € of 1100 million annually for a production of 900 000 tons (example 1). • Underestimation of biomass 5% can potentially lead to a loss of € 126 million in lost income for a production of 900 000 tonnes (Example 2). In addition, it can get lost revenue associated with that has an untapped production potential .Estimation of biomass can get negative consequences for farming, harvesting / processing and sale of salmon and trout. Error Estimation of biomass can lead to: Increased feed conversion due to suboptimal feeding (underfeeding / overfeeding). Suboptimal medication which at worst can cause the breeder must repeat treatment and potential development of resistance among lice at sub-terapautic doze of drug. Loss of potential revenue in sales units resulting from over-estimation, wrong size distribution and the estimation of biomass. The two main economic consequences of improper estimation of biomass are expressed through sub-optimal use of feed lower sales. Examples of calculations of economic consequences for, over-and underestimation was calculated, and is available in Deliverable 6.4 Economic Viability Study and Benchmarking). The FishScan project will develop a system for continuous remote monitoring of weight, growth performance and size distribution of fish for use within aquaculture enclosures, primarily in sea cages. The aim is to achieve an accuracy of 99% in size measurements of fish larger than 1kg and 97 % accuracy for fish at 100g. This will provide the farmer with more accurate data on mean weight and size distribution of the sampled fish from the farmed fish population. The higher accuracy in the FishScan system than in the current state-of-the-art has a potential to assist a reduction from the current ±5% to ±3% error in biomass estimates. The system will be designed to be easy to use to allow workers to spend their time on other tasks. Thus FishScan will be an important tool to enhance economic and ecological sustainability in the European aquaculture industry. Research is still on-going, however the current indications show that FishScan can reach a deviation below ±2% as compared to today’s technologies which have a ±5% deviation. Whilst FishScan did not as yet manage to attain all the initial aims till now, the consortium still feels that with additional research, these objectives can be met. The higher accuracy obtained by using the FishScan system compared to current state-of-the-art can facilitate a reduction of error in biomass estimates from the current ±5% (annual € 91 M loss) to ±3% (annual € 50 M loss) By implementing the improved FishScan system into their everyday farming operations, the European cage culture industry has the potential to release up to € 41 M annually in economic benefits. FishScan: A tool for reducing the costs of suboptimum production for the average fish farmer In most countries permission from governmental authorities is required to farm fish. Such permissions very often put restrictions on the amount of fish that can legally be produced at any site. In Norway, aquaculture permissions for salmonidaes include a production limit at 780 tonnes per license (a fish farming company may have several license). According to law, the standing stock cannot exceed this limit at any time during the production cycle. The standing biomass has to be reported to the authorities on a regular basis by the farmer. The estimations presented are based on production within one Norwegian license, where typically 1,200- 1,300 tons of salmon or trout are produced annually. Currently the average error in biomass estimation is ±5% (Aahus, 2009). By overestimating the biomass (+5%), the farmer loses € 156,900 per license. By underestimating the biomass (-5%), the farmer loses € 22,750. The reason for the large difference between under and over-estimation is that it is easy to calculate the cost of excess feeding, but difficult to estimate the cost of unused production potential as a result of underestimation (i.e. the loss due to unused growth potential in a fish stock that receives too little feed for optimum growth). Even a small reduction of the error in biomass estimation from ±5% to ±3% will provide the farmer with improved revenue of either € 74,390 (over-estimation) or € 6,500 (under-estimation). The higher accuracy in the FishScan system than in the current state-of-the- art has a potential to assist a reduction of error in biomass estimates from the current ±5% to ±3%. Thus the impact of using a FishScan system represents € 40, 445 p.a. per each fish farm license. This corresponds to potential economic benefits from the project of € 41 M annually in increased profits for the European aquaculture industry. Return on the investment for FishScan end-users As estimated above, the impact of using a FishScan system will be an average of € 40,445 p.a. per license (approx. 4 sea cages). Thus, if a fish farmer buys one FishScan unit for one cage, the purchase will start to give return on the investment after 2 years (the calculation includes the € 18,000 invested in the FishScan unit and an additional € 1,000 in annual maintenance costs). The FishScan system should therefore be a lucrative investment for the aquaculture industry. The market for FishScan The increase in European aquaculture production over the last decade has been entirely due to seawater fish farming. It is also sea-based fish farming which has the most obvious and largest potential for further expansion. As the world population level is growing and freshwater is becoming an increasingly scarce and valuable resource, output from freshwater aquaculture will be limited. The opportunity now exists for the aquaculture industry to engage in large scale sea-based production as a consequence of the limitations of both capture fisheries and land-based farming (EC, 2009) and (Stevenson, 2008). Primary markets for FishScan Approximately 80 species are currently cultured in sea cages. Atlantic salmon (Salmo salar) accounts for some half (51%) of all cage culture production and rainbow trout (Oncorhynchus mykiss) nearly one tenth (9%). Atlantic salmon is currently the most widely cage-reared fish species by volume and value (Tacon, 2007)The primary markets for the FishScan innovations are the European salmonidae, sea bass and sea bream cage culture industry. The Atlantic salmon and rainbow trout cage culture industry in the northern parts of Europe (e.g. Norway, UK, the Faeroes, Ireland and Denmark) and the sea bass and sea bream cage culture industry in the Mediterranean (e.g. Greece, Turkey, Spain and Italy) will be target markets for the FishScan system. Total production for these species in the above mentioned European markets was about 1,320,000 tons in 2008 (Lem., 2009), (FEAP, 2008) and (FAO, 2009). The estimates for the salmonidaes are based on the average cage being 120 metres in circumference with a fish density of 14 kg/m3, giving an average of 320 tons in each operational cage. However, there are always a number of cages temporally out of operation, either awaiting re- stocking of smolts, or on-hold for splitting of the stocked fish from one cage to several other cages (e.g. because of high fish density or need for size grading, etc.). By including these temporary inactive sea cages, the average biomass held in one salmonidae sea cage is 280 tons.122 The estimated number of cages is conservative (e.g. there has been reported to be 10,200 cages in Chile in 2003) (Rojas, 2007). Additionally, there are estimated to be some 1000 - 2000 large sea cages in the Mediterranean Sea for European sea bass (Dicentrarchus labrax) and gilthead sea bream (Sparus aurata), which are the main farmed fish species in this area. In 2004 there were 310 licensed large cage farms in Greece and another 345 sea cage farms in Turkey (nearly all used for sea bass and sea bream culture) (Cardia, 2007). Secondary markets for FishScan The easiest non-European markets to penetrate are the salmonidae cage culture sectors in Canada and Chile. This is because the main fish species reared in sea cages in these areas are Atlantic salmon and rainbow trout. Thus the FishScan system can be used without the need for additional adjustment to adapt the system for these species. However, it is believed that there will be extra costs related to adapting FishScan for use in Coho salmon production. Most of the production in Canada and Chile are destined for markets in the Americas. It is therefore not considered to be a threat to the European aquaculture industry to introduce the FishScan system into these markets subsequent to the introduction in Europe. The FishScan system is also expected to be adapted for use on other marine species in cage culture, such as Atlantic Bluefin tuna, Atlantic cod, meagre, etc. The system may also be introduced for use in large scale land-based aquaculture systems. Impact for the participating SMEs The FishScan partner Storvik Aqua AS has built up a substantial customer base of biomass monitoring systems (www.storvik.no). These customers are now asking for solutions that are more suitable for use in modern aquaculture operations. Storvik Aqua AS provide a wide range of technical equipment for the aquaculture industry and has sold approximately 480 fish monitoring units of the biomass frame model. The frame model was developed in collaboration with the FishScan partner Fibula. The current unit price of such state-of-the- art monitoring systems is €14,000 (Romstad, 2009). The new FishScan fish measurement system will be more accurate than currently available products and will be designed and optimised for continuous monitoring of farmed stock from the first time the juvenile fish is stocked into sea cages until the time of harvest. The estimated unit price for the FishScan system is € 18,000. A minimum of 1,000 FishScan units are estimated to be sold to the European salmonidae cage farming industry, adding up to € 18 M in sales. These estimates are based on previous experience in the fish monitoring equipment market, and the fact there is a strong interest amongst farmers for a fish size monitoring system that can provide data of high accuracy in both small and large sea cages. The main salmonidae markets in Norway, Denmark and the Faroes will be reached through Storvik Aqua AS’ office in Norway. Storvik Aqua AS also has a subsidiary company in the United Kingdom (Storvik Ltd. Scotland) that will specifically target the markets in Great Britain and Ireland. Additionally 300-500 FishScan units are estimated to be sold on the Mediterranean markets (sea bass and sea bream). The FishScan partner AquaBioTech Ltd., which is located on Malta, will target the Mediterranean markets. The potential sales for FishScan on the European primary market therefore add up to € 27 M with a 25% market penetration. However, further 400-600 units are expected to be sold on the secondary markets, especially to the salmonidae aquaculture industry in the Americas. Storvik Aqua AS has a subsidiary company in Chile (Storvik SA Chile) that will be used as a sales agent for specifically targeting the markets in Chile. The SMEs in the FishScan project expect to have a free cash flow of € 405,000 from sales after the 7th year the product is on the market. This equals accumulated sales of € 11.7 M and a net present value (NPV) € 1.3 M. NPV is an amount that expresses how much value an investment will result in. This is done by measuring all cash flows over time back towards the current point in present time. As the NPV method resulted in a positive amount, the project was undertaken. Commercialisation and time-to-market for the FishScan system is estimated to be approximately 1-2 years post project on the primary European markets. The time to optimise the system for other species and introductions into the secondary markets is believed to be 2+ years post project. Some comments received in the end-user survey further substantiate the consortiums’ sales expectations. • “It’s important to guarantee accuracy for all size ranges. Our company would invest for 1% additional accuracy.” Salmon Producer, Chile • “We could save 1,5mill USD per year if we were feeding correctly” – Norwegian salmon producer • “We could 2-3% of feed costs with optimal feeding” – Norwegian salmon producer • “Sales price can go decrease by 6 euro cents due to inaccurate biomass estimations” – Norwegian salmon producer • “Optimal biomass estimation could give us 10-25% increased growth” – Norwegian salmon producer • “Inaccurate biomass estimation can lead to keeping the fish too long in water” – Israeli seabass/seabream producer • “Optimal biomass estimation could give better control with sea lice due to shorter time of circulation” – Norwegian salmon producer • “It’s difficult to know when to start therapeutic treatments if you don’t have accurate biomass estimations” – Israeli seabass/seabream producer • “Inaccurate biomass estimation will cause water quality problems and increased oxygen demand” – UK sea bass producerList of Websites:The address of the project public website is www.fishscanproject.com4.2 Use and dissemination of foregroundA plan for use and dissemination of foreground (including socio-economic impact and target groups for the results of the research) shall be established at the end of the project. It should, where appropriate, be an update of the initial plan in Annex I for use and dissemination of foreground and be consistent with the report on societal implications on the use and dissemination of foreground (section 4.3 – H).The plan should consist of:Section A This section should describe the dissemination measures, including any scientific publications relating to foreground. Its content will be made available in the public domain thus demonstrating the added-value and positive impact of the project on the European Union. Section BThis section should specify the exploitable foreground and provide the plans for exploitation. All these data can be public or confidential; the report must clearly mark non-publishable (confidential) parts that will be treated as such by the Commission. Information under Section B that is not marked as confidential will be made available in the public domain thus demonstrating the added-value and positive impact of the project on the European Union.