AN ANALYSIS OF YOLO VARIANTS FOR EFFECTIVE CLASSIFICATION AND DETECTION OF INDIVIDUAL PALM TREES IN UAV IMAGERY

Abstract

Advancements in modern information technology, particularly artificial intelligence, have reduced the need for human interaction while increasing productivity, accuracy, and dependability. Intelligent irrigation systems, along with customized fertilizer and pesticide applications, can greatly enhance the health and productivity of trees. For precise tree classification, sophisticated Deep Neural Network based techniques as You Only Look Once (YOLO) model and Convolutional Neural Networks (CNNs) can be utilized. These methods enable farmers to accurately identify and monitor a wide range of tree species, improving the management of growing conditions and health monitoring. Ultimately, this results in higher output and lower expenses. In this study, we have focused on Individual Palm Tree Identification. Various iterations of the YOLO method, including versions 5, 6, 7, and 8, have been thoroughly trained over a data-set of multiple variety of images of Palm trees and other trees taken from urban as well as plant estates through UAV to classify and detect individual tree canopies present in an scene captured by drone and further contrasted each YOLO algorithm version's performance. YOLO v8 version performed at its best with highest 85.7% mAP, 83.0%precision and 79.2% recall. For researchers and professionals in this field, this research enhances decision-making by providing valuable insights into which YOLO models are best suited for identifying palm trees in UAV datasets.