The launch of the TerraSAR-X satellite marked the beginning of the availability of radar systems with spatial resolutions up to one meter and a revisit time of 11 days. These two fundamental features have allowed, in the last six-seven years, the growth of new perspectives in the monitoring of urban environments. Traditional classification algorithms based on image segmentation can now be surpassed by the exploitation of the frequent multi-temporal SAR images, which lead to a pixel-by-pixel analysis. Such improvement could be exploited in order to integrate ISPRA's land use monitoring techniques that, up until now, have been mainly based on optical data, ground monitoring nets and inventories. This thesis therefore focuses on the implementation of a classification and change detection algorithm characterized by three classes of targets (buildings, vegetation, roads and squares) aimed at urban monitoring applications. In order to classify the pixels, some mathematical tools such as mean, kurtosis, entropy and the correlation to a seasonal sinusoidal trend have been implemented on the time series of the pixel's amplitude. Buildings time series are characterized by high amplitudes and relatively low entropy; roads and squares time series by low amplitudes, high entropy and high kurtosis; vegetation time series by low amplitudes, high entropy and high correlation to the seasonal trends subjected to a positive phase-shift. Additionally, the estimation of the interferometric coherence of the phase allowed a first rough distinction between the metropolitan area and the surrounding countryside. During a later stage, the classification algorithm was integrated with a Bayesian change detector. Thanks to this further analysis, it became feasible to classify the targets before and after the change occurred, monitor the growth of new buildings and identify known events. Finally, two peculiarities that appeared during the development of the thesis became object of investigation. The first one is the presence of buildings that scatter with a sinusoidal trend in time whereas the second one regards the stealth nature of some skyscrapers.
Il lancio del satellite tedesco TerraSAR-X ha segnato l'inizio della disponibilità di sistemi radar aventi tempo di rivisitazione di undici giorni ed in grado di produrre immagini caratterizzate da risoluzioni spaziali di un metro. Queste due fondamentali caratteristiche hanno permesso, negli ultimi sei-sette anni, la nascita di nuove prospettive per il monitoraggio urbano. I tradizionali algoritmi di classificazione, basati sulla segmentazione della singola immagine, possono adesso essere superati grazie all'utilizzo delle frequenti immagini SAR che hanno portato ad una analisi pixel-per-pixel. Tale miglioria può essere utilizzata per integrare le tecniche di monitoraggio dell'ISPRA che, ad oggi, sono principalmente basate su dati ottici, reti di monitoraggio a terra ed inventari. Lo scopo di questa tesi è quindi quello di implementare un algoritmo di classificazione e change detection indirizzato ad applicazioni di monitoraggio urbano. Per poter classificare i pixel sono state effettuate delle analisi statistiche quali media, kurtosi ed entropia ed è stata stimata la correlazione tra la serie temporale dell'ampiezza del pixel ed un andamento stagionale sinusoidale. È stato rilevato come le serie temporali di pixel di edifici siano caratterizzate da ampiezze alte ed entropia relativamente bassa; quelle relative ai pixel di strade e piazze da ampiezze basse, entropia e kurtosi alte; quelle relative ai pixel di vegetazione da ampiezze basse, entropia alta e correlazione con andamento sinusoidale avente sfasamento positivo alta. Inoltre la stima della coerenza interferometrica ha permesso una prima, grossolana distinzione tra area metropolitana e campagna circostante. Durante una fase successiva, l'algoritmo di classificazione è stato integrato con uno step detector baiesiano. Questa ulteriore analisi ha reso possibile la classificazione dei target prima e dopo lo step, il monitoraggio della crescita di nuovi edifici e l'identificazione di eventi noti. Infine, due caratteristiche che hanno attirato l'attenzione durante lo sviluppo della tesi sono state oggetto di ulteriori analisi: la prima è il fatto che alcuni edifici scatterano in modo sinusoidale nel tempo; la seconda riguarda la natura stealth di alcuni grattacieli.
Monitoring the urban environment with multitemporal SAR data
ROSSETTI, GAIA
2013/2014
Abstract
The launch of the TerraSAR-X satellite marked the beginning of the availability of radar systems with spatial resolutions up to one meter and a revisit time of 11 days. These two fundamental features have allowed, in the last six-seven years, the growth of new perspectives in the monitoring of urban environments. Traditional classification algorithms based on image segmentation can now be surpassed by the exploitation of the frequent multi-temporal SAR images, which lead to a pixel-by-pixel analysis. Such improvement could be exploited in order to integrate ISPRA's land use monitoring techniques that, up until now, have been mainly based on optical data, ground monitoring nets and inventories. This thesis therefore focuses on the implementation of a classification and change detection algorithm characterized by three classes of targets (buildings, vegetation, roads and squares) aimed at urban monitoring applications. In order to classify the pixels, some mathematical tools such as mean, kurtosis, entropy and the correlation to a seasonal sinusoidal trend have been implemented on the time series of the pixel's amplitude. Buildings time series are characterized by high amplitudes and relatively low entropy; roads and squares time series by low amplitudes, high entropy and high kurtosis; vegetation time series by low amplitudes, high entropy and high correlation to the seasonal trends subjected to a positive phase-shift. Additionally, the estimation of the interferometric coherence of the phase allowed a first rough distinction between the metropolitan area and the surrounding countryside. During a later stage, the classification algorithm was integrated with a Bayesian change detector. Thanks to this further analysis, it became feasible to classify the targets before and after the change occurred, monitor the growth of new buildings and identify known events. Finally, two peculiarities that appeared during the development of the thesis became object of investigation. The first one is the presence of buildings that scatter with a sinusoidal trend in time whereas the second one regards the stealth nature of some skyscrapers.| File | Dimensione | Formato | |
|---|---|---|---|
|
2014_10_Rossetti.pdf
accessibile in internet per tutti
Descrizione: Testo della tesi
Dimensione
13.71 MB
Formato
Adobe PDF
|
13.71 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/96502