UAV Test Data Generation Method based on CycleGAN
Kunyu Gao, Jinbo Wang, Bin Wang, Ruixue Wang, Jiao Jia · 2021 8th International Conference on Dependable Systems and Their Applications (DSA) · 2021
Unmanned aerial vehicles (UAV) have developed rapidly in recent years. With the popularity of artificial intelligence, UAV has been closely related to the military, economic and life fields. However, there are still security concerns with intelligent software. If these hidden dangers appear in reality, they will cause economic losses and even endanger people’s lives. Thus, intelligent software must be tested and verified before they can be put into service. To ensure the effectiveness of testing, we need to generate high-quality, large-scale and low-cost datasets. However, due to weather, technical conditions and other factors, the existing datasets are difficult to meet the demand. Accordingly, we propose a novel method to generate dataset for UAV software testing based on Cycle-Consistent Adversarial Networks (CycleGAN) in this paper. In addition, We conduct experiments and verification on the real dataset of Google Maps. The results show that even if the dataset is disturbed, the images we generate are still of high quality.