MACHINE VISION SYSTEM FOR LARVAL FISH REARING
Most marine fish hatch from mm-size eggs irrespective of adult size, and suffer prodigious mortality (rates of 70-90% of the brood) in the first weeks of their lives. This mortality makes the rearing of larval (very young) fish costly, and an impediment for introducing new species to the growth cycle. We propose an automated system, based on machine vision, capable of providing real-time, large-scale and on-site characterization of the main parameters that indicate larval quality. The system will help growers on decision making in the growth process.
While it is possible to measure the conditions in the growth tank (temp, oxygen, etc.), growth protocols are “blind” to the cohort’s response. Today there is no way of tracking the animals’ development, activity and well-being.
An automated system capable of providing real-time, large-scale and on-site characterization of the main parameters that indicate larval quality. The system uses machine vision to provide reports of larval development, detecting the occurrence of morphologic aberrations, while quantifying feeding performance and activity levels of larval fish in rearing tanks. The system will enable growers to make informed decision on feeding
regimes and quantities, terminate failing cohorts sooner rather than later, and help develop ways for efficient feeding and quality control.
Machine vision system for larval fish rearing in aquacultures such as:
• Onshore aquaculture, as in the case of fish tank, ponds, aquaponics or raceways, where the living conditions rely on human control such as water quality (oxygen), feed, temperature.
• Inshore aquaculture, where the cultivated species are subjected to a relatively more naturalistic environment.
• Offshore aquaculture, where the species are either cultured in cages, racks or bags, and are exposed to more diverse natural conditions such as water currents (such as ocean currents), diel vertical migration and nutrient cycles.
Laboratory proof of concept:
• A dedicated hardware prototype was developed;
• Proprietary big data was collected (hours of underwater high-speed videos);
• A novel machine learning analysis tool, relying on anomaly detection, is being developed.
Provisional patent application
Eyal Shamur, Miri Zilka, Tal Hassner, Victor China, Alex Liberzon and Roi Holzman (2016). Automated detection of feeding strikes by larval fish using continuous high-speed digital video: a novel method to extract quantitative data from fast, sparse kinematic events. Journal of Experimental Biology (2016) 219, 1608-1617 doi:10.1242/jeb.133751. https://journals.biologists.co...