Concepts

How PCOT works (and why)

PCOT was originally designed to help scientists and engineers analyse PanCam data and produce useful secondary data. It acts downstream from the Rover Operations Control Centre (ROCC) on images which have already been processed to some extent, and is a successor to ExoSpec ¹. As such, its primary purpose is to generate relative reflectance images and spectral parameter maps, although it will also be able to produce spectra from regions of interest. Indeed, it should be flexible enough to perform a wide range of unforeseen calculations.

PCOT can also handle many other kinds of data. It is particularly suited to processing multispectral images with uncertainty and error data, and can currently read PDS4 and ENVI formats, alongside more common RGB formats which can be collated into multispectral images.

Of paramount importance is the verifiability and reproducibility of data generated from PCOT. To this end, a PCOT document is a data product which can be shared between users, which also fully describes how the data was generated from the initial import to the final output. Users are encouraged to exchange PCOT documents in addition to, or instead of, the generated images or data.

The Graph

To achieve this, a PCOT document manipulates data in a graph - a network of nodes, each of which takes some data, works on it, perhaps displays some data, and perhaps outputs derived data. Technically speaking, this is a "directed acyclic graph": each connection has a direction, going from the output of one node to the input of another, and there can't be any loops.

As an example, consider that we might want to overlay some kind of spectral parameter map, converted to a colour gradient, over an RGB image (note: I'm not a geologist, I'm a software engineer, so perhaps this is a very artificial example). One way to do it might be this:

We can see the graph in the panel on the right, showing each node as a box with connections to other nodes (ignore the green numbers, they just show how many times each node has run - it's a debugging aid!) The colours on the input connectors at the top and the output connectors at the bottom show the types of the values accepted and produced by those connectors, as shown on this page.

Here's what each node in the graph is doing:

• The input 0 node reads input number 0 into the graph. The inputs are set up separately from the graph, and can be multispectral images or other data (e.g. housekeeping) from outside PCOT.
• The rect node lets the user draw a rectangle on the image to define a region of interest. Images can have many regions of interest and several different kinds are available.
• The node with 4 inputs a,b,c,d is an expr node, which calculates the result of a mathematical expression performed on each pixel. The node is showing the expression it is running: a$671 / a$438. This will read the bands whose wavelengths are 671nm and 438nm in the node's a input, and find their ratio for every pixel. The result will be a single-band image. Expr nodes can perform much more complex calculations than this.
• The gradient node will convert a single-band image into an RGB image with a user-defined gradient and inset it into the RGB representation of another image - here we are insetting into the input image, using the RGB representation used by that node.

Here is another example, showing a spectral plot:

• The input node again brings a multispectral image into the graph.
• The multidot node adds a number of small, circular regions of interest. Each has a different name and colour, in this case set automatically to just numbers and random colours. Creating the regions is as easy as clicking on the image.
• The spectrum node plots a spectrogram of the regions present in the image for all the wavelengths in that image.

Here I have "undocked" the spectrum node's view to be a separate window for easy viewing. The spectrum can also be saved as a PDF or converted into CSV data. I'm also showing the entire app, including the menu bar and four input buttons.

The Document

A PCOT document is a file which can be shared among users. It consists of

• The inputs - data loaded from sources external to PCOT. These are kept separate from the graph, because you might want to use a different graph on the same inputs, or the load the same inputs into a different graph. There are currently up to four inputs, but this can easily be changed.
• The graph - a set of nodes and connections between them which define operations to be performed on inputs, as shown above.
• The settings - these are global to the entire application.
• Macros - these are sets of nodes which can be used multiple times and appear as single nodes in the graph, although each one has its own "private" graph. Currently very experimental (and largely undocumented).

Quantities

All numerical quantities in PCOT - whether scalar values or pixels in images - consist of three values: nominal value, uncertainty and data quality (DQ) bits. Uncertainty and DQ bitss can be viewed as overlays in the canvas (PCOT's image viewer component).

Uncertainty

The uncertainty is expressed as standard deviation, and is propagated through most operations. Be careful here, however - all values are assumed to be independent. This can lead to grossly incorrect results. For example, we could set up a graph consisting of a node with the value $0\pm1$, feed it into a mathematical expression (expr) node as variable *a), and set the node's expression to $a-a$:

The result, $0\pm\sqrt{2}$, is a consequence of the subtraction assuming that all quantities are independent even when they are clearly the same quantity.

Data quality bits

Each pixel in an image has a set of bits describing its quality in the source data, which are propagated through all operations. Additionally, other DQ bits may be added as the data moves through the graph - for example, division by zero may occur, or an operation whose result is undefined for those values. While DQ bits are flexible, the following are probably stable:

Bit	Meaning
NODATA	no data present (typically for a particular pixel)
NOUNCERTAINTY	Pixel has no uncertainty (again typically for pixels)
SAT	Data is saturated
DIVZERO	A division by zero has occurred
UNDEF	Result of operation is undefined
COMPLEX	Data is real part of complex result
ERROR	Unspecified error

There may also be user-defined or "test" bits for use in development.

Move on to a First Tutorial