Tree detection from high-resolution unmanned aerial vehicle (UAV) images

Abstract

In parallel with the advancement of technology, important developments are taking place in the field of data collection and data processing about the earth, as in many other fields. The data collection stage, which takes a long time with traditional methods, can be completed in a very short time with modern techniques. Satellite images, terrestrial data collection tools, and unmanned aerial vehicles (UAVs) are actively used in studies to determine the characteristics of the earth. UAVs are popular due to their high resolution and fast data collection capacity. Products obtained from UAVs can also be processed based on modern image processing techniques. In this way, it is possible to get meaningful and usable information from the images obtained. The development of fast and easy data collection methods brings along the big data problem. Although long times are spent processing data, success in processing images is low. For this reason, automatic processing of data obtained with modern techniques increases the quality of the result and saves time. In this study, trees were detected over the image of a wooded area obtained with a UAV. The data obtained by the UAV are separated from other objects in the field by object-based classification in eCognition software. In object-oriented classification, the data are grouped at the segmentation stage. Later, segments with similar characteristics were classified according to certain index values. As a result of the study, control data was generated in the eCognition software. Using this data, accuracy analysis was made with the help of an error matrix and the kappa statistic was found to be 78%.