Abstract
Fishery managers, marine biologists, ecologists, and conservationists all rely on accurate estimates of fish population size and composition. Recently, there has been a growing interest in the use of Remote Underwater Video (RUV) methods for fish population monitoring. In particular, stereo Baited Remote Underwater Video (Stereo-BRUV) systems have been shown to be effective for fish population surveys. Stereo-BRUV systems are non-extractive and enable experts to analyse the collected data off-site and asynchronously. However, the increased adoption of RUV techniques has led to a rapid growth in the volume of video data that requires expert analysis. As a result, automating the identification, classification, and counting of fish in underwater footage could significantly improve the efficiency of RUV-based monitoring. This project focused on developing Computer Vision (CV) systems to automate identification and classification of fish from deep sea video collected by the South African Institute of Aquatic Biodiversity (SAIAB). Specifically, we designed deep learning based CV models to perform this task.
Videos 1
Watch presentations, demos, and related content
FishID: Exploring Deep Learning for Fish Classification
Like, comment, and subscribe on YouTube to support the creator!
FishID: Exploring Deep Learning for Fish Classification
Gallery 1
Explore the visual story of this exhibit
Title Image