Binary Semantic Segmentation of Dolphin on UAV Image Using U-Net

Dolphins are mammals that live in both freshwater and saltwater habitats. These creatures are skilled at a variety of acrobatic maneuvers, including turning, jumping, steering rubber boats, and even counting. Dolphins are a symbol of marine tourism in Lovina, Bali because of their special abilities. Although dolphin tracking on Dolphin Tours is still done manually, so technology is required to recognize dolphin sightings. This research employs a binary semantic segmentation approach using the U-Net to learn new information about the detection of dolphin appearance using UAV footage. The dataset used in this study consisted of 1,400 images, of which 1,120 from 80 percent were used for training, 140 from 10 percent for validation, and 10 percent for testing (140 images). The U-Net model results were produced with a mean IoU value of 0.862, a recall of 0.909, and a precision of 0.789 through the adjustment of the hyperparameter training. The results of this study can be used to further investigate how dolphins recognition system might become integrated into UAVs.