Comparative Analyses of Unsupervised PCA K-Means Change Detection Algorithm from the Viewpoint of Follow-Up Plan

Area of Science:

• Remote Sensing
• Data Science
• Image Analysis

Background:

• Supervised methods for Synthetic Aperture Radar (SAR) data analysis are prevalent but have limitations.
• Unsupervised methods offer advantages in computing time, data scarcity, and explainability for trustworthy systems.
• Principal Component Analysis and K-Means Clustering (PCAKM) is a benchmark unsupervised method.

Purpose of the Study:

• To analyze and enhance Principal Component Analysis and K-Means Clustering (PCAKM) for SAR data.
• To reduce algorithm sensitivity to parameter and input image variations.
• To improve accuracy and computation time for change detection applications.

Main Methods:

• Analysis of Principal Component Analysis and K-Means Clustering (PCAKM) algorithms.
• Modification and evaluation of 22 PCAKM configurations using difference images and filtering methods.
• Calculation of error metrics, computing times, and utility functions.

Main Results:

• The modified PCAKM demonstrates reduced sensitivity to input variations and improved accuracy.
• Performance varies with different image characteristics, but overall robustness is enhanced.
• The study provides a foundation for faster, more explainable, and less sensitive change map generation.

Conclusions:

• Enhanced PCAKM offers a more robust and accurate unsupervised approach for SAR data analysis.
• The findings support the development of reliable change detection systems.
• This research addresses a gap in fast, explainable, and sensitive change mapping for follow-up plans.