Semi-automatic Tree Detection from Images of Unmanned Aerial Vehicle Using Object-Based Image Analysis Method

The aim of this study is detection of trees at low cost, fast and reliably through object-based image analysis method by using high-resolution unmanned aerial vehicles data. The study consists of three basic process stages. In the first stage, high-resolution aerial photographs were obtained by using unmanned aerial vehicles. Aerial photographs with approximately 1 cm spatial resolution were obtained from 30 m. flight height with 80% front overlap and 80% side overlap. Orthorectification, mosaicking, georeferencing and filtering processes were performed on the obtained photographs in order to make them ready for the analysis. Orthoimage in three spectral bands (Red, Green, Blue) and digital surface model were generated by processing the aerial images. The second stage of the study constitutes the most important process steps of object-based classification: They are segmentation and classification. In this phase, various parameters such as scale, shape and integrity were used to obtain the best result. Furthermore, accuracy assessments were performed through field observations. In the classification process step, a specified set of rules applied to segmentation result objects; thus, trees and other classes were determined in the study area. The generation of the rule sets was performed using the height information obtained from digital surface model and spectral data collected from orthoimages. Accuracy analysis was performed in the last stage of the study. The number of trees determined in the classification step was compared with the known number of the trees in the field. As the result of comparison, tree numbers and crown widths were found to be consistent. This study tests and implements a method that detects trees at low cost, fast and high accuracy which can be used in ecological studies and specially in determination of tree density.