Cartographic generalization is a problem, which poses interesting challenges to automation. Whereas plenty of algorithms have been developed for the different sub-problems of generalization (e.g., simplification, displacement, aggregation), there are still cases, which are not generalized adequately or in a satisfactory way. The main problem is the interplay between different operators. In those cases the human operator is the benchmark, who is able to design an aesthetic and correct representation of the physical reality. Deep learning methods have shown tremendous success for interpretation problems for which algorithmic methods have deficits. A prominent example is the classification and interpretation of images, where deep learning approaches outperform traditional computer vision methods. In both domains-computer vision and cartography-humans are able to produce good solutions. A prerequisite for the application of deep learning is the availability of many representative training examples for the situation to be learned. As this is given in cartography (there are many existing map series), the idea in this paper is to employ deep convolutional neural networks (DCNNs) for cartographic generalizations tasks, especially for the task of building generalization. Three network architectures, namely U-net, residual U-net and generative adversarial network (GAN), are evaluated both quantitatively and qualitatively in this paper. They are compared based on their performance on this task at target map scales 1:10,000, 1:15,000 and 1:25,000, respectively. The results indicate that deep learning models can successfully learn cartographic generalization operations in one single model in an implicit way. The residual U-net outperforms the others and achieved the best generalization performance.

Cartographic generalization is a problem, which poses interesting challenges to automation. Whereas plenty of algorithms have been developed for the different sub-problems of generalization (e.g. simplification, displacement, aggregation), there are still cases, which are not generalized adequately or in a satisfactory way. The main problem is the interplay between different operators. In those cases the benchmark is the human operator, who is able to design an aesthetic and correct representation of the physical reality.Deep Learning methods have shown tremendous success for interpretation problems for which algorithmic methods have deficits. A prominent example is the classification and interpretation of images, where deep learning approaches outperform the traditional computer vision methods. In both domains – computer vision and cartography – humans are able to produce a solution; a prerequisite for this is, that there is the possibility to generate many training examples for the different cases. Thus, the idea in this paper is to employ Deep Learning for cartographic generalizations tasks, especially for the task of building generalization. An advantage of this task is the fact that many training data sets are available from given map series. The approach is a first attempt using an existing network.In the paper, the details of the implementation will be reported, together with an in depth analysis of the results. An outlook on future work will be given.

Knowledge about land cover is relevant for many different applications such as updating topographic information systems, monitoring the environment, and planning future land cover. Particularly for monitoring, it is of interest to be not only aware of current land cover but of past land cover at different epochs, too. To allow for efficient, computer-aided spatio-temporal analysis, digital land cover information is required explicitly. In this context, historic aerial orthophotos and scanned historic topographic maps can serve as sources of information, in which land cover information is contained implicitly. The present work aims to automatically extract land cover from this data using classification. Thus, a deep learning-based multi-modal classifier is proposed to exploit information from aerial imagery and maps simultaneously for land cover prediction. Two variants of the classifier are trained, utilizing a supervised training strategy, for building segmentation and vegetation segmentation, respectively. Both classifiers are evaluated on independent test sets and compared to their respective two uni-modal counterparts, i.e. an aerial image classifier and a map classifier. Thus, a mean F1-score of 62.2% for multi-modal building segmentation and a mean F1-score of 83.7% for multimodal vegetation segmentation can be achieved. Detailed analysis of quantitative and qualitative results gives hints for promising directions for future research of multi-modal classifiers to further improve the performance of the multi-modal classifier.