Automatic species recognition, length measurement and weight determination, using the CatchMeter computer vision system

Abstract

The collection of biological data on species composition and individual length and weight of specimen has always been an important part of fisheries research. Traditionally, the collected information has been recorded on paper prior to being entered into a computer for analysis. Electronic measuring boards that record length measurements, such as the FishMeter (Øvredal and Totland, 2000), have made the data collection process more efficient and reliable. In this contribution we describe a vision-based catch registration system called the CatchMeter that makes it possible to automatically recognise fish species and measure the length and weight of fish. A camera and light unit is located above a conveyer belt moving at a maximum speed of 1.5 m/s. One ore more images of the fish are taken as it passes the camera and are then analysed by the computer. Both roundfish and flatfish from 5 cm to over 1 m can be recognised and the system can easily be trained to recognise new species of fish. So far the prototype system has been trained to identify 10 different species with a recognition rate of 98.6 %. In a test where the length of the same fish was measured repeatedly, the standard deviation of the length measurement was 3 mm on a 265mm long haddock. Weight measurement has not yet been implemented but we plan to do this by using electronic grader, laser triangulation and/or using length-weight relationships. Provisions for connecting an automatic sorting unit to the CatchMeter will be included to enable sorting of selected individual fish for registration of extended biological parameters. A computer controls all the processes and sorting criteria may be selected from the user interface. With the CatchMeter the sampling process is automated and the capacity for biological sampling will be increased with a reduced need for manpower for measurements and sorting. The new system is initially intended for use on research vessels but it will also be of interest to commercial fishing vessels and at landing sites of fish processing plants. The prototype has been successfully tested on the Norwegian research vessel “G.O. Sars”.