Acoustic tracking of migrating salmon.

Abstract

Annual salmon migrations vary significantly in annual return numbers from year to year. In order to determine when a species' sustainable return size has been met, a method for counting and sizing the spawning animals is required. This project implements a probability hypothesis density tracker on data from a dual frequency identification sonar to automate the process of counting and sizing the fish crossing an insonified area. Data processing on the sonar data creates intensity images from which possible fish locations can be extracted using image processing. These locations become the input to the tracker. The probability hypothesis density tracker then solves the multiple target tracking problem and creates fish tracks from which length information is calculated using image segmentation. The algorithm is tested on data from the 2010 salmon run on the Kenai river in Alaska and compares favorably with statistical models from sub-sampling and manual measurements.