Many state-of-the-art autonomous driving systems require prior knowledge in the form of high definition map data, which provide information about the road and its environment. These data can be erroneous or outdated due to the frequency of updates and because of short- or long-term changes in the road. Any autonomous system that relies on such potentially invalid information needs to detect deviations from the map while driving, by the use of its own, or external sensors, a problem often called online map validation. This paper proposes a novel approach to online map validation, where map data are compared against the readings of onboard sensor measurements by a deep learning classifier. This classifier is based on the Siamese Network architecture, an architecture known from similarity learning. The classifier is trained on data from real world test drives and evaluated on both correct maps and incorrect maps as found in construction sites. Results show that the classifier reaches an F1 score of 89.1%, whereby misclassified scenes mostly stem from the limited variability in the training data and the lacking evidence of construction sites in the input data.