Automated Monitoring of Floating Roof Oil Tanks Using High-Resolution SAR Imagery and Computer Vision Techniques
Yi-Keng Chen (a*), Zheng-Shin Su (b), Shu-Wei Chang (c), Jen-Yu Han (d)

a) Department of Civil Engineering, National Taiwan University, Taiwan
b) Department of Civil Engineering, National Taiwan University, Taiwan
c) Department of Civil Engineering, National Taiwan University, Taiwan
d) Department of Civil Engineering, National Taiwan University, Taiwan

Abstract

Oil tanks are critical national energy infrastructures, and monitoring their fuel capacity is essential, especially in remote or inaccessible regions. This study presents an automated methodology to estimate and monitor floating roof oil tank volumes using high-resolution Synthetic Aperture Radar (SAR) imagery. Unlike fixed-roof tanks, floating roof tanks exhibit observable geometric changes in SAR images due to variable oil levels, making them suitable for remote sensing-based fuel estimation. Leveraging the advantages of SAR-such as all-weather, day-and-night imaging capabilities-we developed a computer vision-based framework incorporating image preprocessing, noise reduction, and geometric normalization to address the inherent speckle noise and imaging distortions of SAR. The methodology includes automated extraction of oil tank features and Hough Transform-based fitting of elliptical or circular arcs to detect tank boundaries, accounting for deformations due to floating roof positions. To enhance accuracy and robustness, we introduce a size normalization procedure and apply constraints to remove secondary reflections that cause false arcs. A U-Net neural network model is further employed for semantic segmentation and feature extraction, serving as a cross-verification mechanism for the automated measurements. Compared to previous approaches that require optical imagery or rely on multi-temporal datasets, our method minimizes data requirements and improves generalizability across different geometric conditions. Experimental results demonstrate the framework^s effectiveness in accurately estimating tank dimensions and fuel levels, even under challenging imaging conditions. This study offers a scalable and practical solution for strategic fuel monitoring in inaccessible regions, with applications in national energy planning, disaster response, and security analysis.

Keywords: SAR, Floating Roof Oil Tank, Computer Vision, Hough Transform, U-Net

Topic: Topic C: Emerging Technologies in Remote Sensing