Remotely sensed above-ground storage tank dataset for object detection and infrastructure assessment

Area of Science:

• Geospatial analysis
• Remote sensing
• Machine learning applications

Background:

• Increasing volume and accessibility of remotely sensed imagery.
• Lack of annotated data hinders automated analysis for machine learning.
• Need for standardized datasets for industrial infrastructure evaluation.

Purpose of the Study:

• To develop a novel, publicly available, multi-class dataset of above-ground storage tanks (ASTs).
• To facilitate large-scale automated analysis of industrial infrastructure using machine learning.

Main Methods:

• Annotation of high-resolution, remotely sensed imagery.
• Development of a dataset including geospatial coordinates, border vertices, diameters, and orthorectified imagery.
• Classification of over 130,000 ASTs into five distinct categories.

Main Results:

• Creation of a comprehensive dataset featuring 130,000+ ASTs across the contiguous United States.
• Dataset includes five labeled classes: external floating roof tanks, closed roof tanks, spherical pressure tanks, sedimentation tanks, and water towers.
• Detailed annotations provided for each AST, enabling diverse analytical applications.

Conclusions:

• The new AST dataset addresses the critical need for annotated data in remote sensing.
• Enables direct use or training of machine learning models for risk assessment, capacity estimation, and infrastructure evaluation.
• Promotes advancements in automated analysis of industrial facilities through publicly available data.