Inventory Management for Christmas Trees Using Unmanned Aircraft

Open Access
Munoz Valdez, Juan Martin
Area of Honors:
Mechanical Engineering
Bachelor of Science
Document Type:
Thesis Supervisors:
  • Henry Joseph Sommer III, Thesis Supervisor
  • Daniel Humberto Cortes Correales, Honors Advisor
  • Object recognition
  • TensorFlow
  • Tree Identification
  • Computer vision
  • Python
  • Inventory Management
Agriculture is essential for our survival, and that is why it should be made as efficient as possible. In this thesis, an algorithm capable of recognizing Christmas trees in aerial images will be developed. The objective is to reduce the amount of work farmers need to do to count the trees and make their inventory management much more informative and efficient. The farmers will use drones to acquire aerial images of the farm. These will then be fed to a prototype program that finds the geographical location of each tree and stores their information. Two different methods were tested, first deep learning and then and a more deterministic approach. The machine learning method was not successful because of many reasons, like the lack of a big data set. On the other hand, the deterministic approach proved to be accurate. It uses thresholding and pattern matching as its main components. The results of this method were then tested using different parameters so that the most efficient configurations could be defined. Once the position of the trees is found in pixels, then the information has to be extracted out from a DEM GeoTIFF image and stored in a CSV file. Finally, the prototype program was developed and specified to fit user needs.