2025 Summer Research Symposium • Joshua Riojas • July 9, 2025

Joshua Riojas

Class of 2026

Major: Electrical Engineering

Mentor: Ben Abbott, PhD, St. Mary’s University

A Convolutional Neural Network for Atmospheric thermal detection in

Unmanned Aerial Vehicles

Birds utilize atmospheric thermals, which are vertical columns of warm air generated by the

uneven heating of the Earth's surface, to conserve significant energy during flight, particularly

over long migratory routes. This natural energy-saving strategy presents a compelling model for

enhancing the endurance of Unmanned Aerial Vehicles (UAVs). This study aims to apply this

biomimetic concept to UAVs to conserve onboard energy and thereby extend the drone's

operational flight time. The primary objective of this research is to develop and validate a

method for a UAV to autonomously detect and utilize atmospheric thermals using a machine

learning model to predict the presence of these updrafts in real-time. A Convolutional Neural

Network (CNN) model was developed to predict whether the UAV is in an atmospheric thermal

based on vertical error derived from its XYZ coordinates. Furthermore, the model incorporates

gyroscope data (roll, pitch, and yaw) to determine the UAV's position and orientation within the

thermal. The quantitative analysis revealed that the CNN model accurately predicted the

presence of atmospheric thermals by correlating vertical error with the UAV's spatial coordinates

and gyroscope data. The successful implementation of the CNN demonstrates that it is feasible

for a UAV to autonomously identify and orient itself within atmospheric thermals using standard

onboard sensors. This step advances energy-efficient UAVs mimicking bird soaring, using

natural updrafts to extend flight time, reducing battery dependence, and enabling long missions

like monitoring, surveillance, and sensing.

Keywords: Unmanned Aerial Vehicles (UAVs), Atmospheric Thermals, Convolutional

Neural Network (CNN), Energy Conservation, Vertical error, Flight

Endurance, Autonomous Soaring, Thermal Detection, Avian Flight