A GAN-Based UAV Platform Complex Weather Image Restoration Technology

With the continuous development of UAV technology, the application of UAV is more and more extensive. In aerial photography, reconnaissance, detection and autonomous driving navigation, the application of UAV is inseparable from the acquisition of clear images. However, in some complex weather conditions, such as foggy weather, windy and rainy weather, etc., the quality of the target image obtained by the UAV imaging system may be affected to a certain extent. In these processes, there will be some unavoidable external factors and hard conditions, which will reduce the quality of the image and make the information that people can obtain from the image become blurred. The data used in most UAV image system training is a data set composed of clear images. In order to solve the image acquisition of UAV in complex weather environment, this paper studies an image inpainting method based on Cycle GAN. This paper uses Jun-Yan Zhu's open source Cycle GAN program to train a repair network model for complex weather images suitable for UAV use.