A simple library for unsupervised Polarimetric SAR change detection

Introduction

This is a simple library for unsupervised Polarimetric SAR change detection. The main function is the **unsupervised_CD** function in unsupervised_CD.py.

The workflow of change detection is:

1. read C3 (covaraince matrix) of the PolSAR data
2. perform boxcar smoothing to ensure non-negative definteness of C3 matrix
3. calculate the pixel-wise distance between two PolSAR images. Availabel distance metrics: Bartlett distance, revised Wishart distance, symmetric revised Wishart distance
4. perform generalized histogram thresholding algorithm to segment the change and unchange pixels

Example

Edit in the if __name__ == '__main__' part of unsupervised_change_detection.py like below

if __name__ == '__main__':
    fa = r'path/to/C3/folder_A'
    fb = r'path/to/C3/folder_B'
    gt = r'path/to/groundtruth/file'
    save_path = r'path/to/save/result/images'
    
    confusion_matrix = unsupervised_CD(fa, fb, save_path, gt,
                        is_print=True, distance_type='Bartlett')

    print(f'confusion_matrix:\n{confusion_matrix}')

The change detection result will be like that below:

time A	time B	change detection result

Requirements

This repo need my another library

License

This project is released under the Apache 2.0 license.