Detection of Archaeological Sites from Aerial Imagery using Deep Learning

In recent years, Deep Learning has proven to be an outstanding tool in the field of computer vision showing promising results in different fields such as the analysis of medical images, obstacle detection for self-driving cars, automatic image caption generation, etc. In the case of Archaeology, the adoption of these methods in the detection of archaeological structures from aerial images has been slower than in other fields. This is an area not widely explored but which seems to have a big potential for the application of Deep Learning methods, given the large amount of airborne data existing. This work presents the results of an approach using 4 different Convolutional Neural Networks (CNN) models based on different architectures and learning methods. Of the models tested, 3 of them correspond to state of the art pre-trained models for which different techniques of transfer learning were used. The fourth one is a CNN architecture developed specifically for this task. The Deep Convolutional Neural Networks used were trained to carry a binary identification task, in this case, to determine whether an image contains any kind of topographical anomalies corresponding to archaeological structures, or not. The case studies were obtained from the southern Baltic sea region of Sweden and Birka and these correspond to aerial images in the visible light range and infrared. The kind of structures present on the images are burials of different shapes corresponding to the Viking ages. By using the Receiver Operating Characteristic (ROC) curve and Area Under the Curve (AUC) as measurement, the selection of the model best suitable for this task was carried out. Additionally, different augmentation techniques were tried including the generation of images using a Deep Convolutional Generative Adversarial Networks. Finally, an ensemble approach was tested combining the results obtained from the models which showed the best results individually in different types of airborne data. With this approach, a sensitivity of 76$% with a specificity of 92% was achieved.Several civilizations have flourished and decayed over the short period of time in which humans have inhabited the third planet of a system which belongs to a yellow dwarf star called Sun. This star has been their source of inspiration, hope, questions, and energy and to praise it, as well as just other natural elements, they have built temples. Many of these constructions have already crumbled, gone forever from the memory of mankind, lost in the haze of ancient times. Some others have withstood the harshness of the weather and other civilizations, remaining in the surface as a remembrance of archaic ages, and as a warning about the fragile nature of human civilization, which sometimes vanishes leaving no trace behind but a pile of rocks. However, there are some others, which have been lost but not forever. Covered by vegetation or new human-made structures they have remained silent, waiting for someone to discover them again and tell the secrets they've kept for their own for so long. It has been almost 350,000 years of human history, and it is now maybe just the beginning of a new era in which the task of detecting the remains of lost civilizations could be delegated, to some extent, to something else than humans: artificial systems with the ability to analyze a large amount of airborne data, looking for terrain anomalies. So far, the methods used for this purpose rely almost completely on the analysis by a specialist. This is a time consuming task which requires highly specialized knowledge and which may involve interpretation biases. These artificial systems, which are based on the use of the most astounding programming paradigms: Artificial Neural Networks (ANNs) and Deep Learning (DL), have already proven to be a reliable tool in the field of computer vision showing promising results in different areas such as the analysis of medical images, self-driving cars, automatic image caption generation among many others. Unlike conventional algorithms, DL and ANN do not have a set of rules predetermined for the system to solve a task since the beginning, instead; given an initial set of data, the system itself finds and recognize the patterns which are useful to make further predictions and analysis. Just as the brain, a single artificial neuron is not capable of all the wonders of which the human mind is capable, but together they can achieve impressive results. In this project, by using different architectures of a specific kind of an ANN model called Convolutional Neural Networks (CNNs), aerial images in the near infrared and visible light ranges were analyzed to detect graves from the Viking ages. Different methods for improving the performance were tried, including the generation of synthetic data (also using CNNs), and the combination of different architectures with different kinds of data in order to achieve better predictions. The improvement and research in this methods are of great importance as nowadays there is more data available than the one which is humanly possible to analyze. The detection of archaeological sites is vital, as it allows the retrieval and protection of these structures which can help in the understanding of many aspects of these ancient civilizations, and in general human history, knowledge which otherwise would be lost forever