This research uses high resolution satellite images in combination with an unsupervised Convolutional Neural Network Autoencoder to identify features that can be used to cluster different built environment typologies. Previous remote sensing research uses ground truth data which for some areas is not available or needs manually labeled training data. This research attempts to circumvent the issue of information scarcity in order to create a methodology that can be applied on any city as long as satellite images are available. From the resulting clusters, clusters can be selected which represent areas with high levels of deprivation which in turn can help identifying the deprived areas.