Improved feed utilisation in cage aquaculture by use of machine vision

[摘要] With the harvesting of fish and other aquatic organisms from naturalwaters having reached its upper limit, aquaculture is vital in providing forthe ever increasing demand for fishery products (Boyd, 1999). Notsurprisingly, aquaculture has seen considerable growth over the lastdecade or more.With the rising importance of aquaculture, there is an increased emphasison cost and reducing of waste for environmental reasons. Therefore,attempts to automate or increase efficiency of feeding are constantlybeing explored.On an aquaculture unit approximately 60% of all costs are for feed;therefore high quality feeding management is essential for all fish farmers.The rainbow trout farm at Jonkershoek Aquaculture Research farm nearStellenbosch currently have a feeding management system which makesuse of traditional hand feeding. Handfeeding is not considered optimal, asthe feed intake or pellet loss is not closely monitored resulting in higheroperating costs.Automation of aquaculture systems will allow the industry to producecloser to markets, improve environmental control, reduce catastrophiclosses, minimize environmental regulation by reducing effluents, reduceproduction costs and improve product quality. The history of automatedcontrol in aquaculture has been brief; most of the systems have beencustom-designed, personal computer systems.A very popular approach for an automated feeding system is to monitorwaste pellets beneath the feeding zone of the fish, with a feedback loopthat can switch off the feeder if this waste exceeds a predeterminedthreshold. Other approaches use hydroacoustics to monitor waste pelletsor demand feeders have also been implemented. These approaches however are not considered optimal as automatic feeders do notnecessarily ensure optimal feed intake. Social dominance using demandfeeders does not allow even feeding distribution among all sizes of fish.In this project it was investigated whether an automated feeding systemcan be developed based on fish feeding behaviour. After facing problemswith poor visibility at the Jonkershoek Aquaculture farm nearStellenbosch, video data were acquired from the Two Oceans Aquarium inCape Town. Since it was a feasibility study, the focus was rather toinvestigate whether a predictive model could be generated for fish feedingbehaviour in a more ideal environment which can form a foundation forfurther research.The well-established multivariate methods of principal components

[发布机构] Stellenbosch University