In recent years, with the development of various GAN architectures, studies on image reconstruction from images have accelerated. It is seen that GAN architectures are used in many different fields from health toe ntertainment sector for many purposes such as image colouring, fake image generation, style transfer and increasing the number of images. In this study, it is aimed to detect fake images created by style transfer using CycleGAN in remote sensing images. Fake and real images are tried to be detected with DenseNet121 , one of the transfer learning models, and voting model, which is an ensemble learning model using this model in feature extraction. It is seen that the proposed model with 98.33% accuracy and 0. 9677 MCC value is better than the 86.10% and 0.7220 values obtained from the classical transfer learning model.