The Stock Assessment Program of the NOAA Pacific Islands Fisheries Science Center is responsible for monitoring fish stocks to help guide management in commercial and recreational fishing. The program is responsible for assessing seven species of bottomfish known as the "Deep 7," which consists of six species of snapper and one species of grouper that are economically and culturally important in the Hawaiian Islands. The program checks and assesses this group by gathering three main data types: Abundance (the amount of fish in the ocean), Biology (mainly observed through life history), and Catch. These three types of data make up what is known as the “A, B, Cs of stock assessment.”
One way that “fishery-independent” surveys collect data is through the use of stereo-video camera systems. Marine scientists are using stereo-video camera platforms to produce high-resolution, species-specific, size-structured abundance estimates without actually taking any fish out of the ocean. The system consists of two low-light video cameras with an 80° diagonal field of view that can image targets in natural lighting to a depth of 300 meters in Hawaiian waters. The cameras are designed to float approximately 10 feet (3 meters) above the seafloor and to record video by pointing at a downward angle of 15 degrees. We chose this configuration because the “Deep 7” species are known to school in the water column at least several meters above the bottom, and are known to favor steep and rocky slopes.
The fishery-independent survey stereo-video camera systems collect video and camera footage to produce millions of underwater images of fish. To assess abundance, we need to annotate these images. But given so many images, it is time consuming for a small team of researchers to annotate them all. Machine-learning algorithms could deal with this issue by annotating all of the images with little to no interference. But the machine-learning algorithms require training data in order to learn how to recognize and identify fish. Humans must analyze and annotate these images to create these training data.
With your help, it may now become possible to analyze and annotate all of these images to produce training data for computer vision algorithms and to provide a more accurate representation of fish abundance in Hawaiʻi!
