Entropy Estimators in SAR Image Classification

Area of Science:

• Earth Observation
• Signal Processing
• Information Theory

Background:

• Remotely sensed data are crucial for environmental dynamics, forecasting, and disaster detection.
• Microwave remote sensing, particularly Synthetic Aperture Radar (SAR), offers advantages over optical sensors due to its atmospheric resilience and active illumination.
• SAR images present challenges like speckle noise, necessitating advanced modeling and information extraction techniques.

Purpose of the Study:

• To evaluate the performance of various Shannon entropy estimators for SAR image analysis.
• To investigate the utility of entropy as a feature for supervised and unsupervised classification of SAR data.
• To propose and validate a methodology for optimizing non-parametric entropy estimators for SAR data.

Main Methods:

• Utilized the G0 distribution family as a suitable model for SAR intensity data, accounting for textural variations.
• Applied parametric and non-parametric Shannon entropy estimators to SAR images.
• Implemented supervised and unsupervised classification algorithms using entropy features.
• Developed and tested a fine-tuning methodology for non-parametric entropy estimators.

Main Results:

• Demonstrated the effectiveness of Shannon entropy estimators as input features for SAR image classification.
• Identified optimal entropy estimation techniques for different SAR data characteristics.
• Successfully applied the developed methodology to real-world SAR datasets, showing improved classification performance.

Conclusions:

• Shannon entropy is a valuable feature for extracting information from SAR data.
• The proposed fine-tuning methodology enhances the performance of non-parametric entropy estimators for SAR analysis.
• These techniques contribute to more accurate environmental monitoring and disaster early detection using SAR imagery.