---

Presenter: Gagan Deep Gutta

Faculty Sponsor: Adrian Jordaan

School: UMass Amherst

Research Area: Environmental Science and Sustainability

Session: Poster Session 4, 2:15 PM - 3:00 PM, Auditorium, A40

ABSTRACT

Monitoring juvenile river herring (Alosa pseudoharengus, Alosa aestivalis) emigration is a top research and management priority, yet existing methods rely on very labor-intensive manual annotation of camera trap footage. Prior work demonstrated that YOLOv5-based neural networks could estimate adult fish abundance from continuous underwater video with 87% accuracy for presence/absence detection and 9.4% counting error compared to human observers. However, that system required over 1,000 hours of development and labeling effort, and has not been tested beyond a single site, and focussed on larger adults. This project advances automated juvenile fish monitoring by implementing YOLOv8, a newer object detection architecture with improved small object detection, anchor-free design, and faster inference speeds. We are training YOLOv8 models on annotated underwater camera trap imagery collected from river herring monitoring sites in Massachusetts. We evaluate model performance by benchmarking detection accuracy, counting precision, and processing speed against prior YOLOv5 results. We further develop an active learning pipeline that prioritizes uncertain frames for human review, substantially reducing the annotation burden that has limited deployment of computational monitoring systems. We hypothesize that YOLOv8 will achieve higher accuracy in detecting small, early-season juveniles while requiring less manual labeling effort. Preliminary results indicate that YOLOv8 can reliably detect juvenile river herring across varying environmental conditions, including turbidity, lighting changes, and dense schooling behavior. This work contributes to scalable, cost-effective monitoring tools that can be adapted across sites and species, supporting data-driven fisheries management and conservation of declining anadromous fish populations.