Marketplace

Building Detection

Detect apartments, houses, industrial buildings and sheds in satellite images.

Description

Building Detection identifies buildings in satellite images. The block can detect buildings of various sizes in images with ground sampling distance (GSD) of 0.55m or less. The output is provided as JSON with details of detected bounding boxes coordinates; and the image showing bounding boxes on the buildings. The block is trained on a data-set obtained from South-Eastern Asia.

The algorithm uses deep learning techniques and CNN-based detection architectures to achieve the computer vision objective. The solution is built in Python and uses Tensorflow at backend as deep learning framework. The algorithm processes satellite images with no restrictions on image dimensions.

The use cases are Urbanization monitoring, Infrastructure monitoring, Urban Planning & Design, Construction.

Note: This block currently can be used with Pléiades Streaming and Raster Tiling, Pléiades Download is currently not supported.

General InformationDescription
Block TypeProcessing for Building Detection
Supported input dataGeoTIFF, PNG or JPEG images. The image is expected to have Ground Sampling Distance (GSD) less than 0.55 m.
Output data formatResultant image with overlayed detection bounding boxes.
Performance0.5 IoU and has detection of 0.5 mAP on satellite images with GSD 0.55 m.

Capabilities

Input Capabilities

raster
up42_standard
format
{
  "or": [
    "GTiff",
    "PNG",
    "JPEG"
  ]
}
resolution
0.5

Output Capabilities

raster
up42_standard
bands
> (Propagated)
dtype
> (Propagated)
format
> (Propagated)
sensor
> (Propagated)
resolution
> (Propagated)
tile_width
> (Propagated)
tile_height
> (Propagated)
processing_level
> (Propagated)

End User License Agreement

View the End User License Agreement conditions in a new tab.