Translator Disclaimer
31 August 2018 Bathymetric data processing based on denoising autoencoder Wasserstein generative adversarial network
Author Affiliations +
Proceedings Volume 10835, Global Intelligence Industry Conference (GIIC 2018); 108350O (2018) https://doi.org/10.1117/12.2503788
Event: Global Intelligent Industry Conference 2018, 2018, Beijing, China
Abstract
In view of the complexity and variability of bathymetric data, the paper introduces a new algorithm named DAE-WGAN to construct sea bottom trend surface. This new model is an alternative to traditional GAN training method, combined with the advantages of Denoising Autoencoder (DAE) and Wasserstein Generative Adversarial Network (WGAN). Firstly, the network structure is introduced in detail, in which the critic (or ‘discriminator’) estimates the Wasserstein-1 distance between the generated-sample distributions and the real-sample distributions, and optimizes the generator to approximate the minimum Wasserstein-1 distance, which effectively improves the stability of the adversarial training. Moreover, the generalized Denoising Autoencoder algorithm is added to train the back-propagation process, having a better ability of dimensionality reduction, which improves the robustness of the whole algorithm. Then, using two different types of bathymetric data (seabed tiny-terrain data and Electronic Nautical Chart data), we had long-time experiments to train the DAE-WGAN till optimality, and got the better sea bottom trend surface. Finally, by comparison with other GAN models (such as InFoGAN, LSGAN), the results show that the proposed method has an obvious improvement in accuracy, stability and robustness, and further illustrate the feasibility of this method in bathymetric precise data processing area.
© (2018) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Ruichen Zhang, Yongbing Chen, Shaofeng Bian, and Duanyang Gao "Bathymetric data processing based on denoising autoencoder Wasserstein generative adversarial network", Proc. SPIE 10835, Global Intelligence Industry Conference (GIIC 2018), 108350O (31 August 2018); https://doi.org/10.1117/12.2503788
PROCEEDINGS
6 PAGES


SHARE
Advertisement
Advertisement
Back to Top