A U-Net Based Segmentation and Classification Approach over Orthophoto Maps of Archaeological Sites

In recent years, remote sensing and deep learning methods have frequently been used to segment aerial images obtained by satellite, unmanned, or manned aerial vehicles. Documentation of the earth's surface is significant in the field of archaeology, as it is in many other fields such as analyzing the earth, urban planning, architecture, and engineering activities. In this paper, orthophotos with ground sampling distance (GSD) in the range of 0-30 cm/pixel obtained from UAV (unmanned aerial vehicle) images in archaeological areas were segmented with an algorithm based on U-Net architecture considering the orthophoto deteriorations, and objects. Over 90% accuracy is achieved for blur, distortion, pixels with missing data, tree, plantation, portable objects (car, people, etc.), fixed objects (building, electricity pole, etc.), road, water and stone classes via the proposed method, while 82% accuracy is achieved for the over exposure class.