Aquaculture defects recognition via multi-scale semantic segmentation

Waseem Akram, Taimur Hassan, Hamed Toubar, Muhayyuddin Ahmed, Nikola Miskovic, Lakmal D. Seneviratne, Irfan Hussain

Abstract: Highlights•Detection of aquaculture net defects such as biofouling, vegetation, and net holes.•A novel multi-scale encoder–decoder topology.•A new loss function, dubbed Lt, can constrain the segmentation networks.•A release of the two highly annotated aquaculture datasets.