| 1 |
<h2>DESCRIPTION</h2> |
| 2 |
|
| 3 |
<em>i.gensigset</em> |
| 4 |
is a non-interactive method for generating input into |
| 5 |
<em><a href="i.smap.html">i.smap</a>.</em> |
| 6 |
|
| 7 |
It is used as the first pass in the a two-pass |
| 8 |
classification process. It reads a raster map layer, |
| 9 |
called the training map, which has some of the pixels or |
| 10 |
regions already classified. <em>i.gensigset</em> will then |
| 11 |
extract spectral signatures from an image based on the |
| 12 |
classification of the pixels in the training map and make |
| 13 |
these signatures available to |
| 14 |
<em><a href="i.smap.html">i.smap</a>.</em> |
| 15 |
|
| 16 |
<p> |
| 17 |
The user would then execute the GRASS program <em> |
| 18 |
<a href="i.smap.html">i.smap</a></em> to create the |
| 19 |
final classified map. |
| 20 |
|
| 21 |
<p> |
| 22 |
This module generates signature files of type "sigset". Use module |
| 23 |
<a href="i.signatures.html">i.signatures</a> to manage generated signature files. |
| 24 |
|
| 25 |
<p> |
| 26 |
For all raster maps used to generate signature file it is recommended |
| 27 |
to have semantic label set. |
| 28 |
Use <em><a href="r.support.html">r.support</a></em> to set |
| 29 |
semantic labels of each member of the imagery group. |
| 30 |
Signatures generated for one scene are suitable for classification |
| 31 |
of other scenes as long as they consist of same raster bands |
| 32 |
(semantic labels match). If semantic labels are not set, it will be |
| 33 |
possible to use obtained signature file to classify only the same |
| 34 |
imagery group used for generating signatures. |
| 35 |
|
| 36 |
<p> |
| 37 |
An usage example can be found in <a href="i.smap.html">i.smap</a> |
| 38 |
documentation. |
| 39 |
|
| 40 |
<h2>OPTIONS</h2> |
| 41 |
|
| 42 |
<h3>Parameters</h3> |
| 43 |
|
| 44 |
<dl> |
| 45 |
|
| 46 |
<dt><b>trainingmap=</b><em>name</em> |
| 47 |
|
| 48 |
<dd>ground truth training map |
| 49 |
|
| 50 |
<p> |
| 51 |
This raster layer, supplied as input by the user, has some |
| 52 |
of its pixels already classified, and the rest (probably |
| 53 |
most) of the pixels unclassified. Classified means that |
| 54 |
the pixel has a non-zero value and unclassified means that |
| 55 |
the pixel has a zero value. |
| 56 |
|
| 57 |
<p> |
| 58 |
This map must be prepared by the user in advance by using |
| 59 |
a combination of |
| 60 |
<em><a href="wxGUI.vdigit.html">wxGUI vector digitizer</a></em> |
| 61 |
and |
| 62 |
<em><a href="v.to.rast.html">v.to.rast</a></em>, |
| 63 |
or some other import/development process (e.g., |
| 64 |
<em><a href="https://grass.osgeo.org/grass-stable/manuals/addons/v.transects.html">v.transects</a>)</em> |
| 65 |
to define the areas representative of the classes in the image. |
| 66 |
|
| 67 |
<p> |
| 68 |
At present, there is no fully-interactive tool specifically |
| 69 |
designed for producing this layer. |
| 70 |
|
| 71 |
<dt><b>group=</b><em>name</em> |
| 72 |
|
| 73 |
<dd>imagery group |
| 74 |
|
| 75 |
<p> |
| 76 |
This is the name of the group that contains the band files |
| 77 |
which comprise the image to be analyzed. The |
| 78 |
|
| 79 |
<em><a href="i.group.html">i.group</a></em> |
| 80 |
|
| 81 |
command is used to construct groups of raster layers which |
| 82 |
comprise an image. |
| 83 |
|
| 84 |
<p> |
| 85 |
<dt><b>subgroup=</b><em>name</em> |
| 86 |
|
| 87 |
<dd>subgroup containing image files |
| 88 |
|
| 89 |
|
| 90 |
<p> |
| 91 |
This names the subgroup within the group that selects a |
| 92 |
subset of the bands to be analyzed. The |
| 93 |
|
| 94 |
<em><a href="i.group.html">i.group</a></em> |
| 95 |
|
| 96 |
command is also used to prepare this subgroup. The |
| 97 |
subgroup mechanism allows the user to select a subset of |
| 98 |
all the band files that form an image. |
| 99 |
|
| 100 |
|
| 101 |
<dt><b>signaturefile=</b><em>name</em> |
| 102 |
|
| 103 |
<dd>resultant signature file |
| 104 |
|
| 105 |
<p> |
| 106 |
This is the resultant signature file (containing the means |
| 107 |
and covariance matrices) for each class in the training map |
| 108 |
that is associated with the band files in the subgroup |
| 109 |
selected. |
| 110 |
|
| 111 |
<p> |
| 112 |
|
| 113 |
<dt><b>maxsig=</b><em>value</em> |
| 114 |
|
| 115 |
<dd>maximum number of sub-signatures in any class |
| 116 |
|
| 117 |
<br> |
| 118 |
default: 5 |
| 119 |
|
| 120 |
<p> |
| 121 |
The spectral signatures which are produced by this program |
| 122 |
are "mixed" signatures (see <a href="#notes">NOTES</a>). |
| 123 |
Each signature contains one or more subsignatures |
| 124 |
(representing subclasses). The algorithm in this program |
| 125 |
starts with a maximum number of subclasses and reduces this |
| 126 |
number to a minimal number of subclasses which are |
| 127 |
spectrally distinct. The user has the option to set this |
| 128 |
starting value with this option. |
| 129 |
</dl> |
| 130 |
|
| 131 |
|
| 132 |
<h2 id="notes">NOTES</h2> |
| 133 |
|
| 134 |
The algorithm in <em>i.gensigset</em> determines the |
| 135 |
parameters of a spectral class model known as a Gaussian |
| 136 |
mixture distribution. The parameters are estimated using |
| 137 |
multispectral image data and a training map which labels |
| 138 |
the class of a subset of the image pixels. The mixture |
| 139 |
class parameters are stored as a class signature which can |
| 140 |
be used for subsequent segmentation (i.e., classification) |
| 141 |
of the multispectral image. |
| 142 |
|
| 143 |
<p> |
| 144 |
The Gaussian mixture class is a useful model because it can |
| 145 |
be used to describe the behavior of an information class |
| 146 |
which contains pixels with a variety of distinct spectral |
| 147 |
characteristics. For example, forest, grasslands or urban |
| 148 |
areas are examples of information classes that a user may |
| 149 |
wish to separate in an image. However, each of these |
| 150 |
information classes may contain subclasses each with its |
| 151 |
own distinctive spectral characteristic. For example, a |
| 152 |
forest may contain a variety of different tree species each |
| 153 |
with its own spectral behavior. |
| 154 |
|
| 155 |
|
| 156 |
<p> |
| 157 |
The objective of mixture classes is to improve segmentation |
| 158 |
performance by modeling each information class as a |
| 159 |
probabilistic mixture with a variety of subclasses. The |
| 160 |
mixture class model also removes the need to perform an |
| 161 |
initial unsupervised segmentation for the purposes of |
| 162 |
identifying these subclasses. However, if misclassified |
| 163 |
samples are used in the training process, these erroneous |
| 164 |
samples may be grouped as a separate undesired subclass. |
| 165 |
Therefore, care should be taken to provided accurate |
| 166 |
training data. |
| 167 |
|
| 168 |
|
| 169 |
<p> |
| 170 |
This clustering algorithm estimates both the number of |
| 171 |
distinct subclasses in each class, and the spectral mean |
| 172 |
and covariance for each subclass. The number of subclasses |
| 173 |
is estimated using Rissanen's minimum description length |
| 174 |
(MDL) criteria |
| 175 |
[<a href="#rissanen83">1</a>]. |
| 176 |
This criteria attempts to determine |
| 177 |
the number of subclasses which "best" describe the data. |
| 178 |
The approximate maximum likelihood estimates of the mean |
| 179 |
and covariance of the subclasses are computed using the |
| 180 |
expectation maximization (EM) algorithm |
| 181 |
[<a href="#dempster77">2</a>,<a href="#redner84">3</a>]. |
| 182 |
|
| 183 |
<h2>WARNINGS</h2> |
| 184 |
|
| 185 |
If warnings like this occur, reducing the remaining classes to 0: |
| 186 |
|
| 187 |
<div class="code"><pre> |
| 188 |
... |
| 189 |
WARNING: Removed a singular subsignature number 1 (4 remain) |
| 190 |
WARNING: Removed a singular subsignature number 1 (3 remain) |
| 191 |
WARNING: Removed a singular subsignature number 1 (2 remain) |
| 192 |
WARNING: Removed a singular subsignature number 1 (1 remain) |
| 193 |
WARNING: Unreliable clustering. Try a smaller initial number of clusters |
| 194 |
WARNING: Removed a singular subsignature number 1 (-1 remain) |
| 195 |
WARNING: Unreliable clustering. Try a smaller initial number of clusters |
| 196 |
Number of subclasses is 0 |
| 197 |
</pre></div> |
| 198 |
|
| 199 |
then the user should check for: |
| 200 |
<ul> |
| 201 |
<li>the range of the input data should be between 0 and 100 or 255 but not |
| 202 |
between 0.0 and 1.0 (<em>r.info</em> and <em>r.univar</em> show the range)</li> |
| 203 |
<li>the training areas need to contain a sufficient amount of pixels</li> |
| 204 |
</ul> |
| 205 |
|
| 206 |
<h2>REFERENCES</h2> |
| 207 |
|
| 208 |
<ul> |
| 209 |
<li><a name="rissanen83">J. Rissanen,</a> |
| 210 |
"A Universal Prior for Integers and Estimation by Minimum Description Length," |
| 211 |
<em>Annals of Statistics,</em> vol. 11, no. 2, pp. 417-431, 1983.</li> |
| 212 |
<li><a name="dempster77">A. Dempster, N. Laird and D. Rubin,</a> |
| 213 |
"Maximum Likelihood from Incomplete Data via the EM Algorithm," |
| 214 |
<em>J. Roy. Statist. Soc. B,</em> vol. 39, no. 1, pp. 1-38, 1977.</li> |
| 215 |
<li><a name="redner84">E. Redner and H. Walker,</a> |
| 216 |
"Mixture Densities, Maximum Likelihood and the EM Algorithm," |
| 217 |
<em>SIAM Review,</em> vol. 26, no. 2, April 1984.</li> |
| 218 |
</ul> |
| 219 |
|
| 220 |
<h2>SEE ALSO</h2> |
| 221 |
|
| 222 |
<em> |
| 223 |
<a href="r.support">r.support</a>, |
| 224 |
<a href="i.group.html">i.group</a>, |
| 225 |
<a href="i.smap.html">i.smap</a>, |
| 226 |
<a href="r.info.html">r.info</a>, |
| 227 |
<a href="r.univar.html">r.univar</a>, |
| 228 |
<a href="wxGUI.vdigit.html">wxGUI vector digitizer</a> |
| 229 |
</em> |
| 230 |
|
| 231 |
<h2>AUTHORS</h2> |
| 232 |
|
| 233 |
Charles Bouman, |
| 234 |
School of Electrical Engineering, Purdue University |
| 235 |
<br> |
| 236 |
Michael Shapiro, |
| 237 |
U.S.Army Construction Engineering Research Laboratory |
| 238 |
<br> |
| 239 |
Semantic label support: Maris Nartiss, |
| 240 |
University of Latvia |