Pléiades Neo Super Resolution

This algorithm improves the spatial resolution of the Pléiades Neo image from 0.3 m to 0.10 m (by a factor of 3). This processing block helps users to have a clearer view of their areas of interest at a much lower cost compared to commercial imagery. In addition, it is helpful to detect changes with better spatial resolution.

The algorithm uses convolutional neural networks (CNN) to produce higher image resolution.

This algorithm supports RGB and NED bands. Users can choose either RGB or NED band as a parameter (“NED” or “RGB” as false/true) to avoid long processing time.

Supported workflows:

• Catalog -> Pléiades NEO (Display) -> Order and get the asset from Storage of UP42
• In project workflows, Processing from Storage -> DIMAP GeoTIFF conversion -> Pléiades Neo Super Resolution

Remark: The algorithm also works with Pléiades NEO (Analytics), but it is not trained on 16-bit images. There might be color differences after applying super-resolution to the Catalog Pléiades Neo (Analytics) 16-bit product.

Processing time:
Without considering preprocessing, it takes ~16 minutes to enhance a 3.92 sqkm (6600 x 6600 pixels) Pléiades Neo image using GPU (Nvidia Tesla K80) provided by UP42.

Output:

• Format: GeoTiff (original image and super-resolved image)
• Image band 3 band (RGB or NED), other bands will be omitted from the output image.

Limitation: Super-resolution cannot express features that are not shown in the original image.

More Information:
Tensorflow is used to build a convolutional neural network (CNN) for super-resolution. Unlike other conventional super-resolution networks, high-resolution and low-resolution images are from different satellites. In the case of Pléiades Neo super-resolution, 0.10 m GSD images are used for high resolution.

Click here to learn more about the block’s documentation.