Instance Segmentation-Based Automated Detection and Thickness Estimation of Oil Spills in Aerial Imagery
Year: 2024
Type: conference-paper
Abstract: An oil spill at sea represents a catastrophic environmental event resulting from the release of oil into marine ecosystems. These incidents pose substantial risks to marine biodiversity, wildlife habitats, and coastal populations, often engendering enduring and widespread repercussions. Cleaning up oil spills is costly due to logistical challenges. The accurate measurement of spill characteristics like the volume, thickness, and area of the spill is crucial before deploying cleanup crews to optimize resource allocation and reduce expenses. The main objective of this research is to use computer vision to detect oil spills and estimate its thickness, helping in decision-making processes to clean up the spill area. The system architecture proposed in this study integrates a drone equipped with a camera module to inspect sea areas and capture images. These images are processed using a deployed computer vision segmentation model to detect oil spills and estimate oil thickness. Predicted results help in decision making via a dedicated application by applying predefined criteria to determine the thickness of the spill, which further help in taking actions for the removal of oil spills. The computer vision model developed in this research could detect and estimate oil thickness with an mAP of 91%. The proposed system in this study uses instance segmentation to detect and segment oil spills in drone footage. This computer vision-based approach accurately identifies and outlines oil spill areas, aiding in the selection of efficient cleanup strategies. Real-time monitoring and assessment capabilities enable quick decision making and effective response measures.
Institutions: Manipal University Jaipur, AISECT University
Language: English
Subfield: Pollution
Field: Environmental Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4405724472