Aerial imaging based sea lion count using modified U-net architecture.

Managing livestock in large manufacturing systems is difficult, mainly in vast areas. UAVs are used to collect images of areas of interest and are rapidly becoming a possible alternative. However, extracting relevant information using good algorithms from images are still scarce. In this paper, we present a method for counting sea lions or other animals using a deep learning model for rough location, colour space control to increase the variance between sea lions and background, mathematical morphology to identify sea lions and gather them individually in clustered groups, and image matching to account for image overlap. This paper describes how to use deep learning to automatically count sea lions from aerial photographs. [ABSTRACT FROM AUTHOR]