Use of Generative Adversarial Networks to Altering Remote Sensing Data
- PDF / 1,792,617 Bytes
- 8 Pages / 612 x 792 pts (letter) Page_size
- 63 Downloads / 250 Views
se of Generative Adversarial Networks to Altering Remote Sensing Data M. V. Gashnikova, b, * and A. V. Kuznetsova aSamara
bIPSI
National Research University, Samara, 443086 Russia RAS—Branch of the FSRC “Crystallography and Photonics” RAS, Samara, 443001 Russia *e-mail: [email protected] Received March 31, 2020; revised July 6, 2020; accepted July 8, 2020
Abstract—The paper investigates the use of generative adversarial networks (GAN) for intentional modification of Earth remote sensing data. A generative neural network that includes a special subnet for object boundary inpainting is considered. The network comprises two GAN: the first completes the object boundary, and the second repaints blank areas. Actual remote sensing data are used to test the generative network under consideration. The exemplar-based Patch-Match algorithm is taken as a reference for comparison purposes. The experimental results allow the conclusion that the approach is an effective tool for the intentional modification of large terrestrial area images in falsification of Earth remote sensing data. Keywords: remote sensing data, image modification, falsification, neural networks, object boundaries DOI: 10.3103/S1060992X20030108
INTRODUCTION The paper investigates the algorithms that enable intentional hardly noticeable modifications of particular parts of images to generate forgery remote sensing data (RSD). The unnoticeable altering of RSD portions is a useful means to conceal important information about them. The methods used in the forgery of RSD can also be employed for detection of the fake data. The popularity of such methods grows as RSD become more available. The reliable concealment of a particular region of an image usually involves the removal of any data within the region and the refilling of the blank using the image inpainting algorithms [1, 2]. The inpainting suggests the filling of initially blank areas of an image so that the resultant picture looks natural and consistent. Today the use of generative adversarial networks (GAN) [3] provides best results [4] in image inpainting. The goal of the present paper is to investigate the efficiency of the use of GAN in image-inpainting techniques for intentional modification of images in forged [5] Earth remote sensing data [6]. A GAN [3, 7] comprises two neural subnets: the generator and discriminator. The generator produces fakes the discriminator detects them. This sort of neural networks also owes its popularity to the fact that after the training of a GAN, the discriminator can be used separately as a means to detect the forgery. 1. BRIEF REVIEW OF AVAILABLE NEURAL-NET SOLUTIONS Recent years have seen few researches that use neural networks, including GAN, to facilitate RSD image generation and completion algorithms. In paper [8] a GAN is employed to remove cloudiness from RSD rather than generate fakes. In [9] a GAN is also used to eliminate overcast, yet only smooth temperature distributions of low resolution are considered. In [10] the attempt is made to use modified GANs t
Data Loading...
