.png)
Working with Fisheries and Oceans Canada we examined an opportunity to integrate AI into an existing electronic monitoring workflow. This project involved the development of a web-based software application to be used alongside electronic monitoring review software to support the use of AI within the catch apportionment step of the reviewer workflow. A bilingual branded experience that could be owned and distributed internally was desired.
We began our user experience design with interviewing electronic monitoring reviewers working within the trawl program. Our aim was to understand the current workflow and opportunities to augment with AI. A step in the review process that relied heavily on human reviewer judgement and skill was the "catch apportionment step". During this step, a reviewer would view the catch coming onboard the vessel and visually estimate the dominant species grouping. Secondary species groupings would also be estimated if present. The proportions were then utilized in a formula to calculate estimated catch weights. Automated catch apportionment, using a computer vision approach to calculate the catch proportions had the potential to be more accurate, more reliable, and faster than human assessment. An additional benefit was improving the overall accuracy of catch apportionment for junior review personnel.
With the intended user experience in hand, we developed some visual guidelines informed by the Government of Canada brand guideline.
Custom branded Interface components library supporting application development
The finalized implementation required less than 2.5 minutes of reviewer time per tow and demonstrated performance equal to, or better than, a human reviewer.
The initial pilot implementation of the user interface saw users easily interacting with the software and performing the required tasks in 3.7 minutes per tow on average. A user feedback survey was sent after the first pilot period of use. With feedback received, we made some additional adjustments to the user interface to change the way video was loaded and to allow a single camera view instead of a multi-camera view. During the second pilot period of use, with these changes implemented, users completed the same required tasks in only 2.5 minutes per tow on average. These results show the power of user experience design in software development for AI-assisted workflows. Fisheries and Oceans Canada also required a bilingual interface. We created the ability for the browser to detect a language preset for Canadian-French and automatically display the translated interface when this setting is detected.
.png)
.png)
Great software in the form of customized fisheries tools can help support Fisheries of the Future to maximize the value of human review time and effort.
.png)