Oh What a Drag! Using Artificial Intelligence to Catalyze Efficient and Cost-Effective Video Review for At-Sea Electronic Monitoring Systems

In the past decade, the feasibility of monitoring at-sea fishing activities using remote electronic monitoring (EM) has been successfully demonstrated in numerous proof-of-concept projects related to this technology. While the cost of collecting high quality imagery and video onboard commercial fishing vessels outfitted with EM has plummeted, the cost of processing and analyzing the large volume of footage generated has not gone down and is viewed as a major impediment to the global use and scaling of EM systems. In recent years, there have been numerous efforts to drive down those costs using Machine Learning and Artificial Intelligence. In these efforts, the overwhelming amount of data is what enables machine learning algorithms to be so successful, with one catch. Most algorithms that can assist in an automated analysis capability fall under the category of supervised classification (e.g. which fish, where was it, how many were there). This requires tedious labeling of data, with the promise of eventual reduction in analysis time. We present here some successes in developing libraries of training data (e.g., fishnet.ai), libraries of algorithms for specific tasks (e.g. OpenEM), and techniques for using semi-automated analysis to accelerate the generation of massive libraries of training data.