Explainable deep learning with novel marine domain metrics for oil spill detection

Résumé

One of the major challenges faced by marine ecosystem and the environment in general is oil spills especially in oil producing areas or areas with crude oil infrastructure. This threatens aquatic life, render the water body and the environment polluted and unsafe. However, accurate and detection could minimise the impact through a timely and effective response. Though the deployment of deep learning for oil spills detection using synthetic aperture radar (SAR) images, have proved effective, nevertheless, lack of interpretability of artificial intelligence models makes it a black-box which reduces the stakeholders’ trust especially in crucial applications such as environmental monitoring. This study demonstrates the application of explainable artificial intelligence (XAI) specifically the deep learning model for oil spill detection. The model integrates the SpillNet, a customised Convolutional Neural Network (CNN) architecture with five XAI techniques and unique evaluation metrics suitable for marine environmental monitoring were introduced. These include the Marine Domain Relevance (MDR) for the quantification of oil spill, False Positive Analysis (FPA) for look-alike discrimination and Domain Alignment Score (DAS); an expert-based checklist with composite metric. Our comprehensive evaluation of 20 representative samples from 1002 SAR images shows that Gradient-Weighted Class Activation Mapping (Grad-CAM) achieves the highest domain alignment score (0.608 ± 0.074). The proposed SpillNet model also achieved s egmentation accuracy (in terms of IoU) of 0.830 (83%) and validation accuracy of 90.5%. Thus, making it the most suitable XAI method for operational oil spill detection systems especially in open-ocean scenarios. The system directly supports several United Nations (UN) Sustainable Development Goals, including the Sustainable Development Goal (SDG) 6 (Clean Water), SDG 7 (Clean Energy), and SDG 14 (Life Underwater), by improving environmental protection through reliable AI-based monitoring systems.

Citer ce document

Accès au document

Ce document est hébergé sur son dépôt institutionnel d'origine.

Auteur(s)

Statistiques

Consultations : 1

Téléchargements : 0