An adversarial learning method to optimize spectral and spatial information in Sentinel-2 satellite data

Wiggers, Bram (2020) An adversarial learning method to optimize spectral and spatial information in Sentinel-2 satellite data. Master's Thesis / Essay, Artificial Intelligence.

Preview	Text MasterThesisBramWiggers.pdf Download (4MB) | Preview
	Text Toestemming.pdf Restricted to Registered users only Download (94kB)

Abstract

Images contain two types of information: spatial, in the horizontal and vertical dimension, and spectral, in the domain of color bands. Multispectral satellite data of the Sentinel-2 satellite contains 12 spectral bands. The shortcoming of this data is the combination of high-resolution bands (10m per pixel) and low-resolution bands (20m and 60m per pixel). This study proposes a deep neural network architecture to perform cross-spectral interpolation. This study proves that information in both the spatial and spectral domain can be exploited to increase the resolution of low-resolution spectral bands. Several deep learning architectures are discussed and tested, and the model is compared to traditional methods to perform cross-spectral interpolation. Results of this study show that the deep-learning model can recreate heavily degraded images better than traditional methods: MSE = 0.082 for the model and MSE = 0.305 for linear interpolation. Because the MSE is susceptible to the loss of spatial features due to blurring of images, a real life application score is introduced. Results show significant increase in classification accuracy when data is preprocessed compared to low-resolution classification. Inspection of the adversarial model shows how adversarial optimization introduces a much more volatile loss function compared to regular L2 optimization. The thesis is concluded with the shortcomings of this research and directions for future research.

Item Type:	Thesis (Master's Thesis / Essay)
Supervisor name:	Schomaker, L.R.B.
Degree programme:	Artificial Intelligence
Thesis type:	Master's Thesis / Essay
Language:	English
Date Deposited:	30 Jul 2020 09:19
Last Modified:	30 Jul 2020 09:19
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/22921

Actions (login required)

View Item