Аннотация:The problem under consideration is the blurring of aerial images of the Earth, which reduces the quality of object detection during remote sensing. The purpose of this study is to improve the quality of object detection in blurred aerial images. This goal is achieved by augmentation of training images. The results of the simulation, conducted on a publicly available dataset of aerial images DOTA, show that an increase in the proportion of distorted images and simultaneous increase in the level of these distortions in the training dataset leads to a rise in the quality metric of the analyzed system. At the same time, there is a decrease in the metric of the boundaries overlap between the regions of interest proposed by the trained model and true objects in the test dataset. The conducted research has shown that choosing the optimal level of distortion of the training data set gives a possibility of developing the most stable system, suitable for diverse tasks of remote sensing.