Adaptive binning and spatial profile partial least squares methods in scanning electron microscopy energy-dispersive X-ray spectroscopy and satellite hyperspectral pansharpening
Sihvonen, Tuomas (2023-12-12)
Väitöskirja
12.12.2023
Lappeenranta-Lahti University of Technology LUT
Acta Universitatis Lappeenrantaensis
School of Engineering Science
School of Engineering Science, Laskennallinen tekniikka
Kaikki oikeudet pidätetään.
Julkaisun pysyvä osoite on
https://urn.fi/URN:ISBN:978-952-412-033-3
https://urn.fi/URN:ISBN:978-952-412-033-3
Kuvaus
ei tietoa saavutettavuudesta
Tiivistelmä
Pansharpening is a data fusion method for image data. Traditionally, images from satellite remote sensing are fused, with one image having a high spatial resolution, but a low spectral one, and the other image vice versa. This methodology has been applied to other image fusion tasks than remote sensing. One of these fields is scanning electron microscopy (SEM) coupled with Energy-dispersive X-ray spectroscopy (EDS). However, as the data structure in these images differs from satellite images, the pansharpening tools need to be developed to reflect this difference.
In this work, two new pansharpening methods have been presented. They both lean on the same core idea of Partial Least Squares (PLS) modeling but are otherwise tailored to work on differently structured images. AB-PLS-DA (adaptive binning partial least squares discriminant analysis) is tailored for SEM microscopy and Energy-dispersive Xray spectroscopy (EDS) image fusion. In comparison, the other developed method, spectral profile partial least squares (SP-PLS), functions well in the satellite imaging domain. This method divides the images into subproblems, facilitating fast calculations.
One of the drivers for requiring high-quality EDS images is to utilize them as textures on photogrammetric 3D models constructed of SEM images. This kind of application demands tens of images. Recording one high-quality EDS image can take minutes; thus, acquiring an amount suitable for texture generation can take hours. With the pansharpening, fewer EDS images in lower resolution can be recorded to achieve a good texture.
From SEM-EDS data, the AB-PLS-DA method offers top pansharpening performance, additionally providing chemical insight into the sample. Likewise for the satellite domain, SP-PLS gives good results, while being fast, with no need for training or large amounts of data.
In this work, two new pansharpening methods have been presented. They both lean on the same core idea of Partial Least Squares (PLS) modeling but are otherwise tailored to work on differently structured images. AB-PLS-DA (adaptive binning partial least squares discriminant analysis) is tailored for SEM microscopy and Energy-dispersive Xray spectroscopy (EDS) image fusion. In comparison, the other developed method, spectral profile partial least squares (SP-PLS), functions well in the satellite imaging domain. This method divides the images into subproblems, facilitating fast calculations.
One of the drivers for requiring high-quality EDS images is to utilize them as textures on photogrammetric 3D models constructed of SEM images. This kind of application demands tens of images. Recording one high-quality EDS image can take minutes; thus, acquiring an amount suitable for texture generation can take hours. With the pansharpening, fewer EDS images in lower resolution can be recorded to achieve a good texture.
From SEM-EDS data, the AB-PLS-DA method offers top pansharpening performance, additionally providing chemical insight into the sample. Likewise for the satellite domain, SP-PLS gives good results, while being fast, with no need for training or large amounts of data.
Kokoelmat
- Väitöskirjat [1123]