Why do we need to enhance satellite images?
The use of Image enhancement is to improve the quality and pixels of the images. This is particularly important in the context of enhancing low resolution satellite images to better extract land features like buildings.In this blog we have used Inverse distance weighting (IDW) on Sentinel satellite data to extract building footprints.
Re-sampling satellite images
IDW on satellite images is an effective re-sampling technique for improving resolution. To improve spatial resolution we can use use statistical mean, median or value estimated using IDW. Mean has better applications in re-sampling elevation data. Median is more useful in removing noise in post processing steps after we compute some indices. IDW is better to improve the visual quality of the images.
This is the image that we are using now to test this approach. We have downloaded this image from USGS earth explorer. As we can see the image has some built up area. IDW is applied on this raw image is then simple edge enhancement is carried out. On doing this the images look like this.
Once we do this we will be able to extract building from this image. But the output quality depends on the technique that we use to extract buildings. From the same image we can get two different outputs by using different convolution filters. But in both images we can see the buildings clearly.
High accuracy convolution
Approximate faster version
Application in change detection
Once we choose satellite data and then apply suitable convolution we will be able to use that for change detection. Change detection needs satellite data of more than one timestamp. In the images below we have taken such images and compared them. To arrive at locations where change has happened we use indices such as building index or vegetation index. These indices help us to classify images, define thresholds and then identify changes. We need not manually check the imagery for change, our code can do that task if trained properly with some pre-fed data.
Before vegetation is cleared
During construction of building
Publicly available satellite data can be downloaded by any registered user of USGS earth explorer. Landsat satellite has a pixel resolution of 30 meter which means we can visualize land features which are more than 30 meter by 30 meter. Sentinel satellite has a pixel resolution of 10 meters.Both these can be resampled and can be used to identify built up areas for decently large buildings.