Recognition and counting of oil palm tree with deep learning using satellite image

Abstract

One of the challenges in oil palm plantation is to accurately the number of palm trees in a unit of area. Knowing the number of trees will contribute to better managerial and operational tasks for fertilization. This requires a method that can be used to calculate oil palm accurately in a quick fashion. This study aims to discuss the method of identifying and counting oil palm trees using a deep learning system based on the YOLO algorithm to process the images captured by a satellite from Google Earth. The system prototype has been successfully built and tested to recognize oil palm trees and to count them in a case study area IPB-Cargill Oil Palm Education Garden, commonly called IPB-Cargill Oil Palm Teaching Farm, is in Singasari, Jonggol, Bogor West Java. Based on the experimental results, the used training dataset using YOLO with 2500 step iterations with a loss value of 0.6 obtained an accuracy values of 85.6%, 98.9% for precision, and 86.6% for recall. This shows that the model is adequate to correctly recognize and to count palm tree objects.