Classification of Biofouling Rates in Underwater Images Using Transfer Learning

This study proposes a method for classifying the biofouling rate on hull surfaces using underwater images captured by the camera of a remotely operated underwater vehicle. Underwater hull images for training are directly collected and classified, and image augmentation techniques are applied to improve classification accuracy. Additionally, well-known pre-trained models are fine-tuned using transfer learning. In experiments using a test dataset, the average classification accuracy and single-image classification time of the method were 95.43% and 4.04 milliseconds, respectively. This performance level confirms that the proposed method is applicable to biofouling-rate classification.