[PERSEO] - Introduction

Prisma hyperspectral image Enhancement for Revealing cultural heritage Sites from Earth Observation (PERSEO)
PRISMA hyperspectral data exploitation for the identification of unknown, buried cultural heritage sites and monuments

[PERSEO] - Gallery

[PERSEO] - Tab

 

 

 

Sub-surface or hidden Cultural Heritage sites can be discovered through Earth Observation (EO) data from a variety of sensors (e.g., hyperspectral, multispectral, radar) by identifying and analysing anomalies or traces on bare soils, crops or vegetation that could be connected to the presence of archaeological deposits under them. Buried heritage contexts in fact can limit the growth and development of vegetation or determine a change in the colour and/or humidity level of bare soils above them, leading to an alteration of their spectral characteristics that can be identified through EO spectral images.

The main goal of PERSEO project is to ascertain the suitability of PRISMA hyperspectral data for applications in the cultural heritage domain, namely it aims to set a benchmark in the use of such data for the automatic detection of undiscovered heritage sites. Therefore, it will: i) advance the use of hyperspectral data in the cultural sector and the information extraction from the imagery’s spectral content compared to the possibilities offered by multispectral data; and ii) define a clear pipeline for the use of Machine Learning approaches for automatic image analysis in cultural heritage domain as well as in other comparable contexts.

 

 

PERSEO project will i) develop novel (and improve existing) super-resolution techniques tailored to hyperspectral PRISMA data to increase their spatial resolution; ii) design an unsupervised deep learning architecture for anomaly detection to automatically identify sub-surface cultural heritage sites using unlabelled hyperspectral data; iii) implement a new Artificial Intelligence software library to carry out the detection task.

PERSEO will trial the development of a robust image fusion and super-resolution technique, suitable for strong noise and different noise types, which exploits two well-known properties of hyperspectral images: low-rank and non-local similarity of image patches. The project will also introduce a new unsupervised ML-based method for automatic trace/anomaly detection, which will exploit the availability of a large number of unlabelled data in PRISMA products and address a lack of a priori knowledge, currently the two main obstacles to the development of an efficient automatic object detection technique in the AI domain.

  Results & Dissemination

  News & Communication


 

Principal Investigator: Arianna Traviglia

Affiliated Researchers:    Marco Fiorucci
                                             Paolo Soleni
                                             Marina Ljubenovic
                                             Homa Davoudi

[PERSEO] - Acknowledgments