Author: Mustafa Emre Döş, Murat Uysal
Publishing Date: 2019
Volume: 1 Issue: 1
Today, with the rapid progress of technology, detailed information is obtained by using tools on different platforms about the environment we live in. Images obtained from air and satellite platforms are used in the production of spatial data. These images are used in many areas, from vegetation detection to natural disasters and urban planning. High resolution images in remote sensing data make it easier to detect artifacts. However, in addition to these advantages of highresolution data, the need for human skill and intervention in the classification process is one of the biggest challenges today due to its comprehensive detail content. To date, many classification approaches have been introduced to reduce human factor. Due to the hardware and software tools previously insufficient in computer technology, subjects such as machine learning and deep learning which are accepted as the foundations of artificial intelligence have not been studied much. However, today it has increased in popularity thanks to improvements in graphics processors and software. The performance and deep learning approach in recent studies is more promising than machine learning. In this study, classification is made by using deep learning algorithms which will be an alternative to existing classification methods. The International Photogrammetry Society and the Remote Sensing Society (ISPRS) Vaihingen data were used to test the algorithm. In the data set, 5 detail classes have been selected as ground, building, tree, vegetation and vehicle. The algorithm for these details is on average 99% and the lowest 95%. According to the results of this study, it is seen that deep learning algorithms are a good alternative for automatic classification of distance learning data.
Key Words: Remote sensing, Classification, Deep Learning, Orthophoto