| dc.description.abstract | The preservation and discoveries of ancient structures is an integral part in the understanding of earlier civilisations. The ever increasing speed of urbanization has lead to the loss of crucial information from our distant past. As a consequence of this, the experimentation with faster and more automated methods to detect these structures has seen an increase.
In this thesis we explored one of these methods. We have used the architectures of high performing convolutional neural networks which have been pre-trained on the ImageNet dataset. These pre-trained models have then been fine-tuned to classify qanats or fortresses on satellite and airborne imagery of areas in the Middle East. In the end, the models classify smaller segments of an image, and label them if they are predicted to contain those ancient structures. We then present the results from the models which showcase their performance. | |
