The problem of CFAR detection of thermal anomalies is discussed in this paper for multiple-band, non-homogeneous, non-Gaussian scenario. Data from 4- and 11 μm bands are projected onto a new coordinates system provided by the decorrelating Principal Component Analysis. A robust PCA is obtained by using the Minimum Covariance Determinant estimator for the covariance matrix that acts by strongly reducing the influence of thermal anomalies. A statistical validation has been carried out through a large bulk of data testing that the first and the second data component well fit a Gaussian and a Log-Normal distribution, respectively. Thus the first component directly satisfies the Location Scale property required for a CFAR detection, while for the second component the same property may be satisfied after a logarithmic transformation. A CFAR detection is applied to projected data and results of the two detectors are combined into a fusion block. Thanks to independence of uncorrelated data the two single detections can be combined with an AND or OR rule, and the overall false alarm probability is the product or the sum of corresponding per-channel probabilities. The results obtained in both cases are compared with the standard NASA-DAC-MOD14 product as a benchmark.

CFAR detection of fire events in non-homogeneous non-Gaussian background

Di Bisceglie M;Galdi C;Ullo S
2012

Abstract

The problem of CFAR detection of thermal anomalies is discussed in this paper for multiple-band, non-homogeneous, non-Gaussian scenario. Data from 4- and 11 μm bands are projected onto a new coordinates system provided by the decorrelating Principal Component Analysis. A robust PCA is obtained by using the Minimum Covariance Determinant estimator for the covariance matrix that acts by strongly reducing the influence of thermal anomalies. A statistical validation has been carried out through a large bulk of data testing that the first and the second data component well fit a Gaussian and a Log-Normal distribution, respectively. Thus the first component directly satisfies the Location Scale property required for a CFAR detection, while for the second component the same property may be satisfied after a logarithmic transformation. A CFAR detection is applied to projected data and results of the two detectors are combined into a fusion block. Thanks to independence of uncorrelated data the two single detections can be combined with an AND or OR rule, and the overall false alarm probability is the product or the sum of corresponding per-channel probabilities. The results obtained in both cases are compared with the standard NASA-DAC-MOD14 product as a benchmark.
CFAR detection; MODIS; Principal Component Analysis; robust estimation; thermal anomalies
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: http://hdl.handle.net/20.500.12070/9533
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 2
  • ???jsp.display-item.citation.isi??? ND
social impact