Abstract

Object detection in overhead imagery is of great importance in computer vision. xView is one of the largest publicly available datasets of overhead imagery. Because limited amount of data/images is available for training, the performance of a typical object detection model is expected to be poor without enough training data. In this paper, data augmentation methods by changing/perturbing some of the properties of the images such as changing the color channel of the object, adding salt noise to the object, and enhancing contrast are applied to the xView dataset. Performance evaluation of object detection using YOLOv3 model and augmented data has been carried out. The results demonstrate that the effectiveness of the data augmentation methods depends on both the specific method and the object classes.