The CatchMeter, developed by Scantrol, the University of Aberdeen, IMR and MATCON, is a computer vision system that can automatically recognise the species of a fish and register its length along with other extended information on colour and shape.
The system is currently installed on the IMR research vessel “G.O. Sars” and has been tested on three cruises and operated by non-expert users on a further two. In total more than 30,000 fish and 25 species have now been successfully analysed by the machine.
For trained species the system has been shown to offer a classification accuracy of 98.6% for 10 species and can be easily trained to recognise more as sufficient data becomes available. The minimum length of fish that can be measured is 50mm and, as images can be stitched together, the upper limit is >1.5m. The fish length is measured with an average standard deviation of 3mm.
Fish travel along an opaque blue, fast moving conveyor belt, through a computer vision system and can be presented in any position or orientation, against a complex background containing water, blood and other debris. The capacity of the system is 8,000-40,000 fish per hour with a conveyor belt speed of 1.5m/s.
So far, 25 species of fish have been recognised by the CatchMeter.
During IMR's ecosystem cruise in the Barents Sea in August, several improvements to the system have been implemented and tested:
• Higher resolution images (1392x1024)
• Software functionality and user friendliness improved
• Opaque blue conveyor belt used and backlight system removed
• Length measurement improved and tested thoroughly
• Images of fish used to train system in previous cruises used to classify fish caught in current cruise
Upcoming work includes integrating the CatchMeter system with the FishMeter system onboard, enabling the two to work together and store data in the same database. The organisations involved see the CatchMeter as being the machine that will capture and store length and weight information of all fish, and then it will sort out the fish that need further analysis on a FishMeter. This sorting will be based on stratification rules set by the researchers.
Volume measurement of fish by laser scanning is under development for integration into the system, offering the possibility of weight estimation.
