Building Detection Filter

Building Detection Filter detects newly built buildings with two time-different SAR images from Sentinel-1.

Automatic change detection on SAR images has been a very difficult task because of its noisy structures. This block uses a deep learning algorithm to achieve overwhelming performances on SAR images.

Supported Workflows

• Sentinel-1 L1C (GeoTIFF) -> SAR Building Detection Filter

Important: Please note that the input data block Sentinel-1 L1C (GeoTIFF) is currently not available, which affects the usage of this block. An alternative workflow based on Sentinel-1 L1C (SAFE) will be supported soon.

General Information	Description
Block Type	Processing
Supported Input Data	Sentinel 1 L1C (GeoTIFF)
Input parameters	Set time = null, time_series with different two times for older and newer image (see example for time here), and limit = 1. Set orbit_direction = 'DESCENDING' or 'ASCENDING' in order to compare with imageries with the same direction. Check with “Dry Run” whether two images are selected properly.
Output Data	GeoTIFF
Resolution	identical to the input
Performance	Prediction Accuracy: 0.99 / IoU: 0.35. This algorithm is optimized for detecting large buildings and aggregated residential housings. In some cases, changes on crop fields are detected, which will be improved in the next version.