DroneGIS | Analyzing drone imagery to teach AI concepts to students

Written by
Reading time:
min
Data Processing & Custom Development
October 4, 2021

Students learn about the potential of AI using DroneGIS (prev. ATLAS) to make aerial data analysis more effective and efficient through hands-on experience with the technology.

Customer: Joseph Cerreta, Ph.D., Associate Professor, College of Aeronautics, Department of Flight, Worldwide Campus, Embry-Riddle Aeronautic University, USA

Industry: Academic study of AI


Summary:

  • Students learn about the potential of AI to make aerial data analysis more effective and efficient through hands-on experience with the technology, providing an understanding for application in future business or educational tasks.
  • Flying a UAS is another important educational component of the project, teaching students to plan missions for UAVs and execute flights.
  • During the May 2021 semester, the student team demonstrated 97% ground truth accuracy for the test project.


For over 25 years, the Worldwide Campus of Embry-Riddle Aeronautical University (ERAU) has been providing students around the world with quality asynchronous learning. Embry-Riddle students receive an engaged and effective learning experience through various virtual labs, communities, and online learning tools.

To provide students with an experimental playground and hands-on AI experience, the course faculty, Dr. Joseph Cerreta, uses DroneGIS (prev. ATLAS). This AI-powered platform from SPH Engineering allows users to store and analyze geospatial data.


Embry-Riddle Aeronautical University was already providing distant learning long before the pandemic and thus has a strong understanding of the necessary tools and requirements of students,” comments Joseph Cerreta, Ph.D., Associate Professor, College of Aeronautics, Department of Flight, Worldwide Campus, Embry-Riddle Aeronautic University, USA.


The project requires the students to fly a UAS over nearby cemeteries and create orthomosaics picturing headstones. Then students use ATLAS to build and train a custom AI detector capable of locating headstones of various shapes and sizes and distinguishing them from surrounding grass and trees, which change in appearance from season to season. Over the course of three weeks, students learn about the importance of AI to US industry, how the technology works, and its potential to make aerial data analysis more effective and efficient.

Image 1. Drone image of a cemetery taken during spring 2021 by ERAU students using UgCS to control the UAS

The project starts with students flying a UAS over cemeteries located near to where they live. First, they must prepare the mission in a simulator based on UgCS mission planning software from SPH Engineering. Students then practice the flight using the hardware-in-the-loop (HIL) simulation technique, which imitates the planned path and provides an understanding of different flight parameters. When ready to perform the actual flight, the faculty prepares the set-up in the field and enables students to connect remotely to the ground station and control the Mavic 2 Enterprise Dual drone using a Beyond Visual Line of Sight (BVLOS) waiver from the Federal Aviation Administration (FAA).

Image 2. Screenshot image of a flight over a cemetery with students remotely controlling the UAS
Image 3. Setup of ground station when operating UgCS with the students remotely (Idaho, USA)

With the use of Pix4Dmapper, the students process the obtained aerial images of the cemeteries into orthomosaics and import to DroneGIC (prev.ATLAS), where they master the AI detectors. Additionally, the faculty provides two other sets of cemetery orthomosaics with different appearing grass and vegetation surrounding the headstones (e.g., from brown in winter to green in spring). While some headstones are located under bare branches in winter, in spring they are covered by trees full of leaves. Students can then use their own and two provided orthomosaics to build and train the AI detectors to count differently shaped and sized headstones.

Image 4. Cemetery with overlays to identify 16 different sections provided to students as part of detector accuracy assignment

Once the students feel confident about their trained AI detectors, the faculty provides a detector test assignment (one of 16 sections of a different cemetery, in which the ground truth has been verified). The students report on how many headstones their AI detector counted, with the final grade being based on the overall accuracy of their detector. During the May 2021 semester, the student team recorded a 97% ground truth accuracy rate for the test assignment.

Although May 2021 was the first semester to offer the AI project to undergraduate and graduate students, the presented grades were very impressive. This highlights the fact that DroneGIS (prev.ATLAS) is very easy to learn and utilize. I am looking forward to more students participating in the project during the October semester,” Joseph adds.


SPH Engineering supports academic studies as part of its global UgCS Educational program. To become part of the UgCS Educational program, please send your request to ugcs@ugcs.com.

About