Development of a computer vision fish biomass measurement procedure for use in aquaculture

The aquaculture industry requires accurate and timely biomass or inventory information for the optimum use of the capital invested in its facilities and to minimize operation and feed costs; this information is also needed for accounting and financial aspects. However, most present methods for obtaining fish biomass information are stressful to the fish, time-consuming, labor-intensive, subject to operator error and/or expensive, and routinely produce errors of $\pm$15 to 25% (Klontz,1993). As computers are essentially devices for the rapid performance of repetitious number-based tasks, it seemed that an automated method based on computer vision would be particularly suitable for obtaining such inventory information. There have been other recent applications of computer vision to the problem of measurement in the aquaculture and fishing industry, but it appears that the research herein reported is the first instance in which all of the following elements have been combined: (a) Pumping live fish through an imaging chamber, (b) Imaging chamber equipped with a mirror for obtaining three-dimensional object information, and, (c) Capturing the images of the fish with a video camera and processing this image information automatically with a computer. It also appears that this new procedure produces smaller errors in the length and aggregate biomass than any of the other automated computer vision aquaculture biomass applications reported up to the time of this writing. The experimental fish were a batch of 26 coho salmon fingerlings (230 days old) with an average length of 105.5 mm, and an aggregate weight of 496 g. The fish image data was analyzed correcting for, (a) perspective only, (b) perspective and refraction, and (c) lens distortion, perspective and refraction. These analyses permitted the measurement of the individual fish lengths with average errors of 2.98%, 2.45% and 2.39% respectively. The error in the aggregate batch biomass calculated from the lengths obtained by this last procedure was of only 0.14%. The computer vision procedure described appears quite promising as the basis for the development of a practical method for inventory measurement of fish and other aquacultural species.