Marinesitu Inc

Currently, Most Fish Passage Monitoring at Hydroelectric Facilities Is Performed Manually With Trained Fish Counters Reviewing Up to 24 Hours of Video Recordings Every Day. This Video Is Recorded at Fish Counting Facilities With Viewing Windows Into Fish Ladders. at Facilities Without Fish Counting Windows, Video Is Either Collected by Cameras Looking Down From Above the Water Surface or by Using Underwater Cameras. Our Team Has Recently Deployed a Prototype Camera System and Software Package That Uses a Trained Machine Learning (ML) Model to Automatically Detect, Track, and Count Fish at the Bradford Island Fish Counting Station of the Bonneville Dam in Real-time. Initial Results From This Pilot Project Indicate That Model Counting Accuracies Greater Than 95% Are Possible and, If Adopted Automated Fish Counting Could Reduce the Cost of Enumerating Adult Fish Passage by Over 50% (potential Fte Reduction From Greater Than 24 Hours Per Day to Less Than 8 Hours Per Day). the Completion of This Pilot Demonstration in 2023 Puts the Technology Readiness Level (TRL) of the Base Software and Hardware System at 7 and the Goal of This Project Is to Reach TRL 9 by the End of the Second Budget Period. the Overall Objectives for This Project, Which Will Be Completed in Two Budget Periods, Are to Improve Upon Current Fish Passage Monitoring Capabilities by Expanding ML Model Capabilities While Also Lowering the Associated Costs of Monitoring. These Objectives Will Be Accomplished Through a Combination of Software and Hardware Innovations. With Regard to Software, the Existing ML Model Will Be Expanded to Include Classification of Fish Species, Fish Length, Fish Origin (wild Versus Hatchery Raised), and the Presence of Injuries, Which Will Improve the Value of the Data Output (task 1). in Addition, a Second ML Model Will Be Developed That Uses Stereo-imagery to Provide More Accurate (millimeter Resolution) Fish Length Measurements and Improve Individual Fish Tracking in Schools (task 3). on the Hardware Side, Task 2 Will Focus on the Development and Testing of a High-quality Machine Vision Stereo Camera System to Be Used in Fish Counting Stations and Enable the Stereo Model Development of Task 3. Finally, an Underwater Stereo Camera System Previously Developed for Long-term Monitoring (development of This System Was Funded by a Separate Doe Sbir Which Will Be Completed Before This System’s Use for This Project) Will Be Used to Demonstrate All the Prior Developments at a Facility Without a Fish Counting Window (for More Information, Please See Task 4). the Successful Completion of These Objectives Is Expected to Greatly Improve the Accuracy of Fish Passage Monitoring While Simultaneously Enabling Monitoring at Facilities Without Counting Windows and Greatly Reducing the Need for Manual Review. This Has the Potential to Greatly Expand the Number of Fish Passage Facilities That Are Monitored, Thus Increasing Our Ability to Manage Conservation and Commercial Fisheries and Help Them Coexist With Hydropower.