Urban super resolution

Urban super resolution is a variant of generic AI based super resolution algorithms that is optimized for urban locations. This algorithm improves spatial resolution of images with original resolutions from original Ground Sampling Distance (GSD) of 0.55 m by a factor of 4.

Traditional resolution techniques use interpolative methods that can be erroneous. The algorithm performs well for urban locations that have buildings, roads and vehicles. It can be used as an ensemble to ultimately improve the output accuracy of other object detection algorithms. In certain cases where a clear image is unavailable, this algorithm can help provide a higher resolution image thereby providing a much economical alternative to high resolution imagery.

Technical Description

The algorithm uses generative deep learning techniques and convolutional neural network (CNN)-based AI architecture to improve the image resolution. The algorithm leverages TensorFlow and is capable to process satellite images across a multitude of formats GeoTIFF, TIFF, PNG, JPG or JPEG formats, with no limitations on image dimensions.

General Information	Description
Supported input data	The GeoTIFF input image is expected to have Ground Sampling Distance (GSD) less than 0.55 m.
Output data format	The output is the resultant image with improved spatial resolution and the same file format as the input.
Algorithm Training Data Details	The algorithm has been trained using custom built data sets from satellite images obtained from extremely high resolution satellites.

For more information, visit the provider website here.