AI in Ocean Monitoring: Deep Learning for Marine Pollution, Fisheries, and Underwater Robotics

Abstract

The ocean covers more than seventy percent of the Earth's surface, yet remains comparatively undersampled relative to terrestrial environments due to the cost, difficulty, and hazard of direct human observation at scale, particularly at depth. Artificial intelligence has become an increasingly important tool for closing this observation gap, supporting automated analysis of underwater imagery and acoustic recordings, autonomous underwater vehicle navigation, satellite-based ocean remote sensing, and detection of illegal fishing activity from vessel tracking data. This paper surveys the application of deep learning to ocean monitoring, organizing methods into a taxonomy spanning underwater visual and acoustic analysis, autonomous underwater robotics, satellite ocean remote sensing, marine pollution detection, and fisheries stock assessment. We review applications including marine plastic debris monitoring, illegal, unreported, and unregulated fishing detection, coral reef health assessment, marine mammal conservation, and ocean climate monitoring, summarize benchmarks and datasets used in this domain, and present a case study illustrating an integrated satellite and vessel-tracking pipeline for illegal fishing detection. We discuss challenges including the scarcity of labeled underwater imagery relative to terrestrial datasets, the difficulty of operating machine learning systems on power- and bandwidth-constrained underwater platforms, and the vastness of the ocean relative to available monitoring infrastructure, and outline future directions connecting ocean AI with autonomous ocean observation networks and international fisheries governance.