An Unsupervised Change Detection Method Using Time-Series of PolSAR Images from Radarsat-2 and GaoFen-3

Area of Science:

• Remote Sensing
• Geospatial Analysis
• Signal Processing

Background:

• Traditional pixel-based unsupervised change detection methods are limited to same-sensor data and are susceptible to speckle noise.
• Detecting changes across time-series data from different sensors presents significant challenges.

Purpose of the Study:

• To propose a novel unsupervised change detection method for time-series PolSAR data acquired by different sensors.
• To enhance the robustness against speckle noise and improve the accuracy of change detection.

Main Methods:

• Utilized omnibus and R test statistics to calculate difference images from time-series PolSAR data.
• Employed a Generalized Statistical Region Merging (GSRM) algorithm for speckle noise suppression.
• Applied a Generalized Gaussian Mixture Model (GGMM) for final time-series change detection map generation.

Main Results:

• Successfully detected time-series changes using PolSAR data from different sensors (Radarsat-2 and Gaofen-3).
• Demonstrated significant suppression of speckle noise compared to traditional methods.
• Achieved improved overall accuracy and Kappa coefficient in change detection.

Conclusions:

• The proposed method effectively addresses the limitations of traditional change detection techniques.
• It enables reliable change detection from multi-sensor time-series PolSAR data.
• The method offers a robust solution for analyzing land cover dynamics with reduced noise interference.