Synthetic Datasets for ICSC Flagship 2.6.1. "Extended Computer Vision at high rate" paper #1 "Datacube segmentation via Deep Spectral Clustering"

Synthetic Datasets for ICSC Flagship 2.6.1. "Fast Extended Computer Vision"  paper #1 "Datacube segmentation via Deep Spectral Clustering"

It is a preliminary paper for the ICSC Spoke 2 WP6 flagship 2.6.1 "Fast Extended Computer Vision".

Code repository at: https://github.com/ICSC-Spoke2-repo/FastExtendedVision-DeepCluster

Abstract:

Extended Vision techniques are a ubiquitous in physics. However, the  data cubes steaming from such analysis often pose a challenge in their interpretation, due to the intrinsic difficulty in discerning the relevant information from the spectra composing the data cube.<br>
Furthermore, the huge dimensionality of data cube spectra poses a complex task in its statistical interpretation; nevertheless, this complexity contains a massive amount of statistical information that can be exploited in an unsupervised manner to outiline some essential properties of the case study at hand, e.g.~it is possible to obtain an image segmentation via (deep) clustering of data-cube's spectra, performed in a suitably defined low-dimensional embedding space.<br>
To tackle this topic, we explore the possibility of applying unsupervised clustering methods in encoded space, i.e.~perform deep clustering on the spectral properties of datacube pixels. A statistical dimensional reduction is performed by an ad hoc trained AutoEncoder, in charge of mapping spectra into lower dimensional metric spaces, while the clustering process is performed by an iterative K-Means clustering algorithm.<br>
We apply this technique on two different use cases, of different physical origin: a set of MA-XRF data on pictorial artworks, and a synthetic dataset of simualted astrophysical observations.