Exploratory data analysis of visual sea surface imagery using machine learning

Olga Bilousova^1,2*Mikhail Alekseyevich Krinitskiy^1,2

• ¹Shirshov Institute of Oceanology of the Russian Academy of Sciences (IO RAS), Moscow, Russia
• ²Machine Learning for Earth Sciences laboratory, Moscow Institute of Physics and Technology, Dolgoprudny, Russia

Marine litter is an issue affecting all regions of the World Ocean. To address the problem of World Ocean pollution, it is essential first and foremost to develop observation methodologies capable of providing objective assessments of marine litter density and its sources. One of the most accessible yet still objective observation methods is visual imaging of the ocean surface followed by the analysis of the imagery acquired. The goal of our study is to develop a method for analyzing marine surface imagery capable of detecting anomalies, given that some of the anomalies would represent floating marine litter. For this purpose, we apply our algorithm based on artificial neural networks trained within the contrastive learning framework, along with a classifier based on supervised machine learning method for analyzing optical imagery of sea surface. The approach we present in this study is capable of detecting anomalies such as floating marine litter, birds, unusual glare, and other atypical visual phenomena. We explored capabilities of the artificial neural networks we use in this study within two training approaches with subsequent comparison of the results. Within our sampling approach, we propose to utilize the ergodic property of sea wave fields, leading to significant spatial autocorrelation of image elements with a substantial correlation radius.