BoatNet: automated small boat composition detection using deep learning on satellite imagery.

Area of Science:

• Environmental Science
• Remote Sensing
• Artificial Intelligence

Background:

• Accurate tracking of national carbon footprints is vital for meeting Paris Agreement emission reduction targets.
• The small boat segment contributes significantly to global transportation carbon emissions, yet its emissions are poorly quantified.
• Previous methods for assessing small boat emissions relied on assumptions or specialized sensors, limiting their scope.

Purpose of the Study:

• To develop an innovative methodology for detecting and classifying small boats using open-access satellite imagery.
• To address the knowledge gap in quantifying greenhouse gas emissions from the small boat sector.
• To establish a foundation for estimating carbon footprints of leisure and fishing boats.

Main Methods:

• Utilized deep learning algorithms to process satellite images for small boat detection.
• Developed a novel methodology named BoatNet for small boat identification and classification.
• Applied the methodology to analyze small boat populations in three cities in the Gulf of California, Mexico.

Main Results:

• The BoatNet methodology achieved high accuracy (93.9%) and precision (74.0%) in detecting and classifying small boats.
• Successfully identified and classified leisure boats and fishing boats even in low-resolution and blurry satellite images.
• Demonstrated the potential of open-access satellite data and AI for environmental monitoring.

Conclusions:

• Deep learning-based satellite image analysis offers a powerful tool for monitoring small boat activity.
• The BoatNet methodology provides a scalable approach to quantify small boat populations for emissions estimation.
• Future research should integrate operational data to accurately estimate small boat greenhouse gas emissions.