Our new meeting room was bursting with New Zealand machine learning and fisheries experts on Wednesday for a workshop led by David Middleton, chief executive of Trident Systems.
David says the workshop was about gathering people to look at potential machine learning applications in fisheries – specifically video observation.
“At the moment, electronic monitoring gear on fishing vessels collects GPS tracks and video footage, replacing the human observers who gather data onboard the vessels. Analysing that footage is a fairly intensive process, where people on shore have to watch it and make the same types of observations they would make if they were at sea.”
He believes that advances in machine learning and artificial intelligence can streamline that process and produce reproducible data automatically from the footage collected on vessels.
“There’s an opportunity to use the same or similar tools that have been used for other machine learning applications and apply them to the data we have from fisheries. But there is still research and development to be done.
“You can’t naively point a tool at a heap of fisheries data and expect magic things to come out. There’s still a role for data specialists to work with people who understand fisheries information and make it all hang together. That conversation was what the workshop set out to progress.”
In a presentation for the group, Edward explained why he thinks machine learning should augment what people are doing, not replace it.
“You can use machine learning to scale up what people are doing—this makes it easy to integrate into existing workflows, and allows continual training of the artificial intelligence appplication. Keep a human check on any machine learning calssification tasks helps to maintain confidence in the process”
He also noted that machine learning is data-hungry, so good quality data was needed at the outset to get a good outcome.
Trident Systems provides research and development services for fisheries management and has worked with Dragonfly for a number of years to make fisheries data collection and management more efficient.