Analysis of U-Net Based Image Segmentation Model on Underwater Images of Different Species of Fishes

Image segmentation is one of the key tasks for image processing in robotic vision, navigation, virtual reality, and augmented reality. Segmentation in underwater images can help to overcome many challenges in exploring vast marine biological resources and gene banks. This work compares the performance of image segmentation using the U-Net architecture based model for underwater semantic image segmentation on selected five different fish species from the Fish4Knowledge image dataset. Analysis on the effect of different threshold values on refining the predicted segmented mask was also performed.