Road detection in aerial images with deep learning

Deep learning is an approach based on machine learning algorithms that allows modeling and classification by means of learning the properties for unknown data by modeling with the multi-layer artificial neural network on previously known data sets. As a result of the increase in the usability of artificial intelligence technology in the devices we use in our lives, deep learning has emerged and the studies conducted in this sense have continued to increase day by day. Thanks to deep learning technologies, object recognition and analysis of these objects has accelerated the classification studies. The most basic deep learning algorithms were convulsive neural networks and multilayer artificial neural networks. Especially in the field of image processing, deep learning algorithms have been widely used in analysis studies. In this thesis, the analysis of high resolution aerial images with an important place in the field of deep learning by using Convolutional Neural Networks (CNNs) method and determination of roads will be discussed. With this study, it is ensured that the large-size aerial images can be processed and processed in order to achieve the workable image size. This data set was prepared and then the test was carried out with various deep learning network models. Creating data sets can be used in methods of deep learning activities for Turkey is presented. As a result of the estimation of the trained neural network, the geographic data of the road network in the image was obtained by using the algorithm developed by using HAVERSINE and BEARING formulas. As a result of this study, it has been observed that deep learning has yielded successful results in the detection of roads in high resolution satellite imagery.