CSIRO researchers are developing software that uses machine learning (artificial intelligence) to review electronic monitoring, also known as e-monitoring or EM, footage taken on board Commonwealth commercial fishing vessels. Currently, all videos are reviewed by people, but the project will explore how machines could learn to review and process EM footage, improving efficiencies such as speeding up the reviewing process, hence reducing costs to commercial fishers.
EM is used by the Australian Fisheries Management Authority (AFMA) to help monitor fishing activity in a number of Commonwealth commercial fisheries. EM systems comprise multiple video cameras located in various positions on fishing vessels, which record all fishing activity and store location, date and time data at the same time. The cameras minimise the need to have AFMA Observers on board boats, improving AFMA's monitoring capabilities of commercial fishing activity.
Although the machine learning tool is still in development, positive results have already been seen. Current computer software is able to correctly detect fish on video footage with a precision of between 95 and 99 per cent. In the short term, this software has the potential to reduce the size of video files by removing sections with no fishing activity, so reviewers can then focus attention on sections where fish are actually caught. With more development, the software also has the potential to identify different fish species, opening up the door for real-time quota tracking, monitoring of protected species interactions and much more. This will make footage review much faster and more financially economical.
Machine learning in e-monitoring is a very exciting area of development that is advancing at a rapid pace. The technology could be rolled out to all Commonwealth-managed fisheries in the future, helping to maintain Australia's reputation as a leader in sustainable fisheries management.
