Infrastructure Change Detection Pléiades
Coming back soon: currently not available. Detects man-made infrastructure changes across two different images with a high accuracy.
Infrastructure Change Detection is based on SPOT (1.5m) or Pléiades (0.5m) imagery. This module detects changes in man-made infrastructures such as roads, buildings and houses,
earthworks, etc. with a very high accuracy. This can be used for defense monitoring, ISR, urban monitoring, property tax change detection, unpermitted change detection and other use cases.
HyperVerge’s deep learning based change detection algorithms operate on high-resolution Pleiades or low-resolution SPOT imagery and highlight regions where man-made infrastructure changes have occurred with a high accuracy.
Federal: Monitor sensitive areas of interest to detect any changes that occur in the area such as construction of roads, buildings, artificial islands propping, etc. to ensure that the
petabytes of data can be transformed to trusted, tactically relevant and actionable insights.
Local Governments: Understand where unpermitted changes are happening to take appropriate action for property tax, permit violations thereby increasing compliance, revenue and database accuracy. The same modules can also be used for urban planning.
GIS/Mapping Companies: Keep a tab on which areas are changing and which regions need an update in the maps without having to do any manual work. Let change detection do the rest for you.
- Catalog -> Pléiades Display -> Order and get the asset from storage of UP42
- In Projects workflows, Processing from Storage -> DIMAP GeoTIFF conversion -> Raster CRS Conversion (CRS: EPSG:3857) -> Raster Tiling (tile size: 512, match_image_extents = True) -> Infrastructure Change Detection Pléiades
Set the following parameters to run it successfully:
- Data platform: Set two asset IDs for Processing from Storage
- Set CRS to EPSG:3857 in Raster CRS Conversion
- Set tile size in Raster Tiling to width = 512 and height = 512
- Set augmentation factor = 2
- ‘match_image_extents’ in Raster Tiling set to True
Note: This block fails if less or more than two images are given.
Manual QA Process
Please note that this algorithm was optimized for a high recall score. Therefore, a manual QA process is required to filter out irrelevant predictions.
|Supported input data
|Input tile size set to 512x512 pixels. Compatibility with jpg, png, 3 or 4 bands, tif 3 or 4 bands (8 bits RGB), Resolution 2.4 m, Image padding 32
|Output data format
|GeoJSON vector file in geographic projection containing all changes as polygons.
|Algorithm qualified on a large set of data encompassing different types of landscapes worldwide by Airbus R&D and operational team, with a high level of recall performance. Works better with images of the same season, no snow and restrained angles on dense cities and high buildings.