Earth Server Logo

BGS EarthServer - Remote Sensing

Aerial and Satellite Imagery

There are a wide variety of purposes for which BGS uses imagery. These vary with image type, but automatic image processing is notoriously difficult, even with the latest expensive object based image classification systems. We are currently using a single 3 band Landsat image for testing. Examples of the kinds of queries we would like to perform are:

Implementing some of the above queries is fairly complex. For example, an outline algorithm for finding similar areas would take a polygon around a region deemed to represent a certain kind of land coverage (e.g. coniferous forest) and apply steps like those below to find similar pixels within a certain tolerance:

for each pixel {

	for each training class(k) {
		given g{k} = ln[prior_prob{k}]-0.5*D(j)^2- 0.5*ln[|V(k)|]
		step 1. f(k) = exp(g(k))
		step 2. SUM(f(k))
		} /* end for each training class(k) */

	for each training class(k) {
		step 3. if (SUM(f(k))!=0) post_prob(k)=f(k)/SUM(f(k))
	} /* end for each training class(k) */

} /* end for each pixel */

We have not attempted to implement such algorithms using WCPS yet but have used some simple illustrative queries, such as highlighting areas where blue dominates over other colours, to show the principle of trying to highlight the sea at a very crude level. Try these out on our remote sensing client page.