Convolutional Vision Transformer Modeling for Spectrogram Image Processing in the Detection of North Atlantic Right Whales Up-Call

Whale sound detection is an important topic in marine conservation and research. The North Atlantic Right Whale is a whale species whose population is on the brink of extinction. Some of the threats to the North Atlantic Right Whale population at this time are fishing nets or gear that can get caught on whales and whale collisions with ships. To be able to reduce these events, it is necessary to determine the whale hotspot area to be avoided by ships. In this case, the researcher tries to use a spectrogram image processing approach with the Convolutional Vision Transformer model of underwater sound recordings with whale calls as the main parameter in whale detection. The use of the Convolutional Vision Transformer model is a new approach to spectrogram image processing. The Convolutional Vision Transformer model uses convolutional layers combined with Transformer blocks which in theory can speed up the training process. The model can process spectrogram images well, with 97.25% accuracy, 97.26% F1 Score, 97.34% Precision, and 97.25% Recall. With a faster training runtime compared to the Vision Transformer model.