High-precision groundfish length measurement and species identification

The fish processing industry has traditionally relied on manual methods for determining the type and size of fish, a process that is both labor-intensive and time-consuming. Recognizing this significant challenge, this thesis explores the application of advanced computer vision techniques to automate and enhance groundfish length measurement and species identification. For precise length measurement, this thesis employs Segment Anything Model (SAM) and several image processing methods, including skeletonization, convex hull, erosion, and dilation. Additionally, this work implements transformer models for groundfish species identification, achieving a remarkable accuracy of 94%, surpassing previous efforts that utilized the YOLOv8 model on the self-collected dataset. This integrated approach represents a substantial advancement in automating fish processing, significantly improving both efficiency and accuracy. The findings demonstrate the potential of transformer models and advanced computer vision techniques to revolutionize the fish processing industry.