Graphs.

The toolkit produces a variety of graphs from its option panel.

The main types of graph you may wish to test are

 

 Special graph functions (available on all graphs).

1) The scale on the axes can be easily changed, by double-clicking on the x or y axis, and entering a new maximum and minimum value in the dialogue boxes supplied.

  

2) Where a key is shown, then lines, points or bars for any particular data layer can be switched on or off, by clicking the layer's colour square on the key.

 

While a data set is active, and is being displayed, its colour symbol looks like this;

if it's switched off, the colour symbol looks like this;
 

 

 

Histograms

Value histograms are produced from a single selected data layer, for a quick summary of e.g. spectral values.

Multiple histograms are produced from a set of selected layers, and the layers can be switched in or out, using the key, to show one or more sets of values.

 

 

Scatterplots.

These are set up with standard menus, selecting the data layers to use as x and y axes.
Pixels can be selected for query using the mouse. A selected pixel will also be highlighted in any surface or animation which is running.
To select a box of pixels, click the mouse down on one corner, drag to the opposite corner and release.
 
 

 

Interactive scatterplots.

One or more or the fractional data layers (fuzzy memberships of verification layers) can usefully be plotted in an interactive scatterplot, using spectral bands as the x and y axes. For example, selecting four layers of fuzzy memberships, with TM band 1 as the x axis, and TM band 4 as the y axis, produces a graph looking like this.

 

The slider bar on the left is used to set the threshold value, above which pixels are plotted. The percentage value on the right represents the fuzzy membership threshold as a percentage - e.g. at a threshold percentage of 50%, only pixels with memberships above 0.5 are plotted.
At lower thresholds, many of the pixels will contain a little of each of the selected classes, and so will be plotted 3 times, once in each colour. The effect of this can be seen by changing the class on top, using the 'Data layer on top' menu on the right.

The main purpose of the interactive scatterplot is to try and distinguish between spectral confusion and genuine mixing on the ground. For example, the classes 'tile' and 'tarmac' are strongly correlated in their real distributions, because of the real structure of the urban landscape (small patches of tiled roof, surrounded by pavements and roads). However, they may not be spectrally similar. If, within the two selected spectral bands, their characteristic signatures are different, the as the membership threshold is raised, the clusters of 'tile' and 'tarmac' points should pull apart. If, on the other hand, their association is due to spectral confusion, then even at high membership thresholds, the clusters should stay close together.

 

 

Clusterplots.

The cluster plot is another way in which to visualise the distribution of pixels, and their membership values, in just two bands at a time. Selecting the 'Cluster plot' option on the Graph menu will prompt for

Selecting four fuzzy membership layers, and plotting them against TM band 1, on the x -axis, and TM band 4, on the y-axis, gives a cluster plot looking like this;

 
The different landcover classes are switched on and off using the coloured squares on the key. In each case, higher membership is indicated by darker colour. The darker clusters, therefore, represent the cluster centres, for these two landcover classes, in these two particular spectral bands.

NEXT - Fuzzy spectral signatures