Nel contesto dell’elab di dati ipersp il concetto di rango coincide con la definizione recent fornita in lett di vd
Le componenti della base a rango K vengono determinate utilizzando , l2 infinito
Di seguito si riporta in sintesi uno schema a blocchi dell’algoritmo
Si riportano infine i risultati associati alla stima della vd del dato indian pine … in particolare il dato è costituito da 16 classi distinte
Riepilogando il lavoro effettuato … concentrandosi sui vincoli
Signal Subspace Estimation in Hyperspectral Data for Target Detection Applications 2010 IEEE GOLD REMOTE SENSING CONFERENCE Salvatore Resta , Nicola Acito, Marco Diani, Giovanni Corsini Dipartimento di Ingegneria dell’Informazione, Università di Pisa via G. Caruso 16, 56122 Pisa, Italy 29, 30 April 2010 Accademia Navale, Livorno, Italy
The generic sample, or pixel, of the hyperspectral data can be modeled as the combination of a signal contribution and a noise contribution.
The signal is modeled according to the
Linear Mixture Model (LMM) [Stein,02] .
N c x 1
Hyperspectral sensors are characterized by a very high number of spectral bands and a very accurate spectral resolution .
Spectral Dimension Hyperspectral Data Image intensity for a fixed wavelenght Remote Sensing & Image Processing Group Hyperspectral Data Analysis Spectral Signature of the pixel Wavelenght (nm)
Anomaly Detection & Rare Vectors 3/10 Rare Vectors are often spectral components of the target of interest Anomaly Detection (AD)
No a-priori hypothesis about the target is assumed
The goal is to identify those pixels having a spectral signature which is significantly different from the background
Remote Sensing & Image Processing Group
Surveillance of strategically sensible areas
Change Detection in operative areas
Mine Detection in terrestrial and sea environment
Shipwreck survivor location
Applications Rare Vectors Scarcely represented in the observed data Linearly independent on the abundant vectors which address the background
Dimensionality Reduction (DR) Determination of the Virtual Dimensionality (VD), which is the minimum number of spectrally distinct signal sources that characterize the hyperspectral data from the perspective view of target detection and classification [Chang,04]. Rank Estimation Basis Estimation
DR typically includes two distinct steps:
Projection of the original data onto the estimated subspace 4/10
Dimensionality Reduction (DR) goals:
Rare Vectors preservation in Target Detection Applications Remote Sensing & Image Processing Group Dimensionality Reduction Computational complexity reduction Preservation of major characteristics in the observed data
Rare Vector Original Data Energy BIC - PCA RGB image Indian Pine
DR on a case study
MOCA IRVE - SRRE Residual Energy Projection Matrix 7/10 Maximum Value of residual energy Remote Sensing & Image Processing Group Experiments on a case study RESIDUAL ENERGY RESIDUAL ENERGY RESIDUAL ENERGY BIC - PCA MOCA IRVE - SRRE 80000 329 321
Computational load of IRVE – SRRE algorithm is considerably reduced with respect to MOCA algorithm Traditional methods show a tendency to overestimate the subspace rank 8/10 Remote Sensing & Image Processing Group Experiment on a case study & Computational Complexity ITC – PCA MOCA IRVE - SRRE 154 s 690 s 64 s Computational load Indian Pine
[Ste02] D. W. J. Stein, S. G. Beaven, L. E. Hoff, E. M. Winter, A. P. Schaum, A. D. Stocker, “Anomaly Detection from Hyperspectral Imagery”, IEEE Signal Process. Mag., 19(1), 58-69 (2002).
[Ric93] J. A. Richards, X. Jia, Remote Sensing Digital Image Processing, 9, Springer-Verlag, 1993.
[Aci08] N. Acito, G. Corsini, M. Diani, S. Matteoli, S. Resta, “A novel technique for hyperspectral signal subspace estimation in target detection applications ”, Accepted for International Conference on Geoscience and remote sensing – IGARSS, 2008.
[Kuy07] O. Kuybeda, D. Malah and M. Barzohar, ”Rank estimation and redundancy reduction of high dimensional noisy signals with preservation of rare vectors”, IEEE Signal Processing Magazine, vol. 55, Issue 12, Decemder 2007, pp. 5579-5592.
[Cha04] C. I. Chang and Q. Du, “Estimation of Number of Spectrally Distinct Signal Sources in Hyperspectral Imagery”, IEEE Transactions on Geoscience and Remote Sensing, vol. 42, no. 3, March 2004.
[Sto04] P. Stoica and Y. Selen, “Model order selection: a review of information criterion rules”, IEEE Signal Processing Magazine, vol. 21, Issue 4, July 2004, pp. 36-47.
[Rog96] R. E. Roger and J. F. Arnold, “Reliably estimating the noise in AVIRIS hyperspectral imagers” Int. J. Remote Sens., vol. 17, no. 10, pp. 1951–1962, 1996.
Thank you for the attention! Remote Sensing & Image Processing Group References