COASTVISION - Computer vision to expand monitoring and accelerate assessment of coastal fish

CoastVision develops computer vision methods for the efficient processing of video data from coastal surveys. This approach is applied in case studies focused on the detection, species classification, sizing, tracking, and re-identification (re-ID) of individual fish, thereby enhancing the monitoring and stock assessment of coastal fish populations.

Secondary Objectives

1- Improve existing models for the detection and classification of coastal fish to achieve state-of-the-art accuracy, using training and validation data from multiple surveys.
2- Expand the automated processing pipeline to include fish sizing.
3- Deploy a new camera system for the automated collection of data to support re-identification (re-ID), integrated with radio-frequency identification (RFID) technology.
4- Train and validate re-ID models with a target accuracy of more than 90% over short time periods (< 6 months) for Atlantic cod, corkwing wrasse, and ballan wrasse.
5- Apply the CoastVision pipeline to ongoing fish population surveys and generate key data for stock assessments, including growth estimates and fishery selectivity functions.