An Introduction to Remotely Sensed Data Analysis - Sampling Spatial Units for Agricultural Surveys - page 87

To digitally process the images, they must be in a digital format. Images that are
initially in analog format (for example, an aerial photograph) can be converted to
digital format using a process called optical-mechanical scanning. Below, we give a
brief description of some of the imagery commands. These descriptions are also
available by running the command g.manual from within GRASS.
The module r.in.gdal offers a common interface for many different raster
and satellite image formats. Other import modules are available for special cases.
The full map is always imported. GRASS raster/imagery map processing is always
carried out using the present region settings (see command g.region ); in other
words, it uses the extent of the current region and the current raster resolution. The
geocoding of imagery data is a very important step in the analysis. In particular,
GRASS can geocode raster and image data of various types: unreferenced scanned
maps by defining four corner points (see commands i.group , i.target , i.
points , and i.rectify ), unreferenced satellite data from optical and radar
sensors by defining ground control points (see commands i.group , i.target ,
i.points , and i.rectify ), and orthophoto (see command i.ortho.
photo ).
It is possible to calculate some vegetation indices using GRASS. For instance, to
study the vegetation status with the NDVI (normalized difference vegetation index)
derived from multispectral data, the red (R) and the near infrared channels (NIR)
are used as input for simple map algebra (see command r.mapcalc ). The index is
defined as

¼ NIR NDVI ð R = ð NIR þ R Þ Þ: ð 4 : 28 Þ

For unsupervised classification, GRASS has a two-pass procedure. The first pass is
performed by i.cluster , and the second by i.maxlik . Note that both pro-
grams must be executed for unsupervised classification. The clustering algorithm
i.cluster reads through the raster image data, and builds pixel clusters based on
their spectral reflectances. The pixel clusters are categories that can be related to
land cover types on the ground. The maximum-likelihood classifier i.maxlik
uses the cluster means and covariance matrices from the i.cluster file to
categorize each pixel using the class with the highest probability.
The GRASS procedure for supervised classification is very similar, and has two
steps. The first step is performed using i.gensig or i.class , while the second
uses i.maxlik. i.gensig is a non-interactive method for generating the input
for i.maxlik . First, it reads a raster map layer called the training map, which has
some of the pixels or regions already classified. It then extracts spectral statistics
from an image based on the classification of the pixels in the training map, and
makes these statistics available to i.maxlik . Conversely, i.class is an inter-
active program that allows the user to outline a region on the screen and compute
the spectral statistics based on its cells. During this process, an histogram of the
region for each image band is available to the user. In a supervised image classi-
fication, the maximum-likelihood classifier i.maxlik uses the region means and