Incorporating Image Recognition and Machine Learning into the NE Multispecies Groundfish Electronic Monitoring Programs to Quantify Species and Sizes of Discards

· 01/2022 - 02/2024

Lead PI: Pingguo He, University of Massachusetts Dartmouth

Objective: The objective of the project is to reduce the cost of electronic monitoring programs with AI technology that improves the operational efficiency, accuracy, and timeliness of EM discard data for science and management. Specifically, we will develop a new automated “discard chute” with integrated stereoscopic cameras to automatically identify, count, measure, and estimate volume/weight of sub-legal groundfish that is to be discarded in real time. The chute will be an integral part of a EM system which involves monitoring of entire fishing operation, compliance, data transmission, storage, and security, and other requirements.

Methodology: We will design and build a automatic fishing counting and measuring discard chute prototype with high definition cameras, strobe light and control unit. We will build AI and ML algorithms and codes with deep visual detection and length measurement, precision counting, fine-grained fish categorization, and active incremental learning methodology and functionality. We will test the hardware and software, and build image library in laboratory at SMAST and at sea onboard a fishing vessel operated by the collaborator. We will utilize existing image library at NOAA and at SMAST to accelerate learning.

Rationale: All New England groundfish vessels are required 100% industry funded at-sea monitoring starting in May 2022, with an option to propose a NOVA-approved EM program as an alternative to human observers. The “Optimized Retention” EM model requires fisher to sort discards by species into designated discard totes, sub-sample and measure large numbers of fish, then discarded in camera view, imposing operational burdens and additional costs to the video review. An intelligent “discard chute” with integrated cameras and AI system can prevent disruptions to fishing operations, reduce video review needs, and reduce the cost of wireless video transfer and cloud-based storage.