Banca de DEFESA: ROSA JANETH ALPALA

Uma banca de DEFESA de DOUTORADO foi cadastrada pelo programa.
DISCENTE: ROSA JANETH ALPALA
DATA : 28/07/2025
HORA: 16:00
LOCAL: Remoto
TÍTULO:

Entropy-Based Test Statistics for Heterogeneity Detection in SAR Data

PALAVRAS-CHAVES:

Entropy; SAR; Heterogeneity; Gamma distribution; Hypothesis testing; Bootstrap

PÁGINAS: 85
RESUMO:

This work presents a statistical approach to detect heterogeneity in synthetic aperture radar (SAR)
intensity data using entropy-based methods. In SAR data analysis, accurate terrain interpretation
fundamentally depends on distinguishing two key regimes: homogeneous regions, characterized by fully
developed speckle and modeled by the Gamma distribution, and heterogeneous areas, which exhibit
complex scattering behaviors typically captured by the GI0 distribution. While this discrimination is
essential for remote sensing applications, classical parametric tests often fail to address it effectively due
to analytical limitations and numerical instabilities. To overcome these challenges, we propose three test
statistics for detecting heterogeneity in SAR imagery based on Shannon, Rényi, and Tsallis entropy
measures. These tests rely on nonparametric entropy estimators constructed from sample spacings,
avoiding explicit assumptions about the underlying distribution. To enhance their accuracy, particularly in
small samples, we incorporate a bootstrap bias-correction procedure that improves the stability of the
estimators, reduces bias, and lowers mean squared error. The proposed tests are evaluated using Monte
Carlo simulations, assessing size and power under varying sample sizes and numbers of looks. Results
demonstrate that the Rényi and Tsallis-based tests outperform the Shannon-based test by detecting
subtler texture variations while maintaining greater reliability in identifying truly homogeneous regions.
Finally, the methodology is applied to both simulated and real SAR data. We generate p-value maps using
a sliding window analysis, allowing for visual and quantitative assessment of spatial heterogeneity. The
Rényi-based test consistently identifies fine-scale roughness patterns, while the Tsallis-based test proves
more effective in reliably detecting homogeneous areas. Together, these entropy-based tools provide a
robust, interpretable, and fully unsupervised framework for heterogeneity detection in SAR data.

 

MEMBROS DA BANCA:
Presidente - 2224108 - ALEJANDRO CESAR FRERY ORGAMBIDE
Externo à Instituição - SAEID HOMAYOUNI
Externo à Instituição - PAOLO ETTORE GAMBA
Externo à Instituição - ANDERSON A. DE BORBA
Externa à Instituição - FÁTIMA NELSIZEUMA SOMBRA DE MEDEIROS - UFC