Advanced Deep Learning Framework AquaSense for Comprehensive Marine Pollutant Detection Using Sentinel-2 Multispectral Data

Abstract

Marine pollution poses a critical threat to aquatic ecosystems and human health, yet its automated detection and monitoring using satellite data remain significant challenges. Existing methods often focus on single pollutants or binary classification tasks, limiting their effectiveness in real-world, operational settings. The research intends to develop a robust, scalable, generalizable automated system for detecting and monitoring marine pollution by deep learning utilizing upscaled multidimensional satellite images. Used extensive testing on the Marine Debris and Oil Spill (MADOS) dataset. To address these limitations, suggested the Aqua-Sense, an innovative deep-learning architecture that offers a comprehensive method for detecting marine pollutants. AquaSense, leverages high-resolution multispectral data from Sentinel-2 satellites, allowing to detect variety of contaminants such as plastics, oil spills, and other contaminants. The framework is built on state-of-the-art semantic segmentation techniques, which have been optimized for processing complex marine environments. Aqua Sense significantly improves the robustness and scalability of marine pollutant monitoring systems. In MATLAB, it can simulate and evaluate various algorithms and techniques using built-in functions, toolboxes, and custom code. This framework leverages recent advancements in state-of-the-art architectural designs and demonstrates superior performance, outperforming all baseline models mean Intersection over Union (mIoU) (%), F1-score (%), precision (%), Overall Accuracy (%), Cohen's Kappa (%) and Recall (%) metrics. AquaSense offers significant advancements over existing marine pollutant detection methods by providing a comprehensive, scalable, and globally generalizable solution.