Sean-Wyn Ng, ’21, Computer Science

Certificate(s): Technology and Society

I worked training computer models to automatically identify fish species in Baited Remote Underwater Video systems (BRUVs). Marine biodiversity is often estimated from underwater video footage, but the manual annotation of fish is significantly time consuming. Automating the annotation of BRUVs would drastically improve the efficiency of marine research. I manually annotated approximately 14,000 images from multiple BRUVs, targeting fish species that occurred the most frequently. The images were then fed into convolutional neural network (CNN) models, which are often used in machine learning for automatic image classification. CNNs have internal parameters that are adjusted based on information contained in the training set and these parameters are later used to identify objects in new images. During my internship, I gained practical coding experience and learned more about computer vision techniques, and I developed time-management skills by organizing a large-scale project. I also have a greater awareness of issues related to marine biodiversity and conversation.