Deep Learning Utilization for Building Identification by Single Shot Detector

Abstract

Deep learning which is one of the developments from machine learning, is being widely used today for the process of detecting an object. The well-known algorithm is the Single Shot Detector. One of its uses is to detect objects in the form of buildings. Indonesia, which is one of the vulnerable suburban countries with building density, can be used as one of the locations for this research. The detection process with the Single Shot Detector begins with data collections in the form of imagery, then creating sample training to teach the model to recognize buildings, finding the most efficient learning rate, until finally running object detection using the Single Shot Detector. The results obtained by the most efficient learning rate are at 0.001, which then the object detection process is carried out as much as 20 epochs. With the learning rate and epoch, the final result for training loss was 6.459 and validation loss of 25.710. This proves that deep learning can be used for object detection in form of building with high resolution imagery.