Dwelling Detection on VHR satellite imagery of Refugee/ IDP Camps using Faster R-CNN

This Bachelor Thesis describes a new method for dwelling detection on Very High Resolution (VHR)-Satellite imagery using Faster-RCNN developed for the BMBF-project HUMAN+. HUMAN+ aims to develop a real-time situational awareness application for efficient migration management to guarantee humanitarian security. The described method and a corresponding workflow are used in a remote sensing module in the input layer of the HUMAN+ application. It analyses VHR-satellite images of refugee camps, finds all dwellings on the image and calculates an estimate of the number of tents (i.e. dwellings) and people in the camp. To find the dwellings, a Faster R-CNN is used. R-CNNs build a special neural network on a regular CNN to use them for object detection. This thesis describes the generation of training data, the training of a Faster R-CNN and the utilization of the trained Faster R-CNN in a dwelling detection application.