Kim, H.D.; Aoki, S.; Oh, H.; Kim, K.H.; and Oh, J., 2021. Seabed sub-bottom sediment classification using artificial intelligence. In: Lee, J.L.; Suh, K.-S.; Lee, B.; Shin, S., and Lee, J. (eds.), Crisis and Integrated Management for Coastal and Marine Safety. Journal of Coastal Research, Special Issue No. 114, pp. 305–309. Coconut Creek (Florida), ISSN 0749-0208.
Understanding fluid mud density and depth can contribute directly to the development of strategies to reduce dredging costs and other problems in coastal waterways. Although echo sounders are most commonly employed to measure depth, each country has different ways to classify the sub-bottom echo character classification. This study aims to train and test the YOLOv3 artificial intelligence model to classify sub-bottom profiles along with the Darknet-53 framework. Two dataset packages with different amount of image data were used to compare the model performance. The mean average precision (mAP), loss, average intersection of union (IoU), and F1 scores were compared to interpret the results between dataset packages. Throughout the study, it was observed that when the number of images used in training was higher, the model training performances improved. The results of this study show that the YOLOv3 model was successfully trained and tested on the sub-bottom profile datasets, resulting in a high accuracy of classification detection.