CUR - The Council on Undergraduate Research

Automatic Detection of Fish Using Airborne Lidar and Machine Learning

Joseph Aist, Kyle Rust, Jackson Belford, Bradley M. Whitaker, Department of Electrical and Computer Engineering, Montana State University, 610 Cobleigh Hall, Bozeman, MT, 59718

In 1994, invasive lake trout (Salvelinus namaycush) were found to be feeding on the native cutthroat trout (Oncorhynchus clarki bouvieri) in Yellowstone Lake, threatening the native species’ existence and the ecological balance in Yellowstone National Park. In September 2004, an experiment exploring the feasibility of using airborne LIDAR for mapping lake trout spawning areas in Yellowstone Lake was successfully conducted. The success of this experiment led to its repetition in 2015 and 2016. However, the algorithms that exist to map fish in the open ocean failed due to the highly cluttered underwater environment of Yellowstone Lake. Therefore, experts manually examined the data, which was not time- or cost-effective. In this work, we applied machine learning algorithms to identify lake trout in Yellowstone Lake using LIDAR data from prior studies. Of all the tested algorithms, Support Vector Machines were most effective at identifying regions of lake trout in Yellowstone Lake. Our results can provide biologists with a map of lake trout spawning sites that can be used to help maintain the ecological balance of Yellowstone National Park. Based on the success of this project, similar machine learning algorithms were tested on data from LIDAR flights over the Gulf of Mexico. The results from the Gulf of Mexico identified regions of interest and areas where fish had not previously been detected from manual inspection of the data. As work continues through the winter semester, the algorithms will be tested on similar airborne LIDAR data from the Oregon coast.    

Additional Abstract Information