Abstract:

Information contained in high-resolution images in many areas such as urban planning, meteorology, forestry, mining, agriculture, geology, landscaping, and land surveying is needed for various purposes. There are dense data in images with high resolution, and automatic or semi-automatic detection of the details requested from this dense data contributes to the work to be done in terms of speed, cost, and accuracy. Automatic detail detection has become faster and easier thanks to advanced detail extraction methods. In this study, high-resolution raw data obtained in the visible spectrum (red-green-blue band) with Unmanned Aerial Vehicle (UAV) were evaluated and automatic road detection was made with an object-based classification approach. With the object-oriented classification method, it has been shown that it is possible to classify according to different object properties such as spatial, spectral, and contextual information. As a result of the study, two different analyzes were made to determine the compatibility of the road network detected by the object-oriented approach from the orthophoto image of the highway with the ground reality. Error matrix based on TTA mask approach was used to check the classification quality, and analysis was performed with an error matrix (Confusion matrix) to investigate its compatibility with the existing map. As a result of the analysis, the kappa statistic was 0.7958, the general accuracy and the manufacturer's accuracy were calculated as 0.7520, 0.8158, respectively.

Keywords: