1i.gensigset(1) Grass User's Manual i.gensigset(1)
2
6 i.gensigset - Generate statistics for i.smap from raster map layer.
7
9 imagery
10
12 i.gensigset
13 i.gensigset help
14 i.gensigset trainingmap=string group=string subgroup=string signature‐
15 file=string [maxsig=integer] [--verbose] [--quiet]
16 Parameters:
18 trainingmap=string
19 ground truth training map
20 group=string
22 imagery group
23 subgroup=string
25 subgroup containing image files
26 signaturefile=string
28 resultant signature file
29 maxsig=integer
31 maximum number of sub-signatures in any class
32 Default: 10
33
35 i.gensigset is a non-interactive method for generating input into
36 i.smap. It is used as the first pass in the a two-pass classification
37 process. It reads a raster map layer, called the training map, which
38 has some of the pixels or regions already classified. i.gensigset will
39 then extract spectral signatures from an image based on the classifica‐
40 tion of the pixels in the training map and make these signatures avail‐
41 able to i.smap.
42 The user would then execute the GRASS program i.smap to create the
44 final classified map.
45
47 Parameters
48 trainingmap=name
49 ground truth training map
50 This raster layer, supplied as input by the user, has some of its pix‐
52 els already classified, and the rest (probably most) of the pixels
53 unclassified. Classified means that the pixel has a non-zero value and
54 unclassified means that the pixel has a zero value.
55 This map must be prepared by the user in advance. The user must use
57 r.digit, a combination of v.digit and v.to.rast, or some other
58 import/developement process (e.g., v.in.transects) to define the areas
59 representative of the classes in the image.
60 At present, there is no fully-interactive tool specifically designed
62 for producing this layer.
63 group=name
65 imagery group
66 This is the name of the group that contains the band files which com‐
68 prise the image to be analyzed. The i.group command is used to con‐
69 struct groups of raster layers which comprise an image.
70 subgroup=name
72 subgroup containing image files
73 This names the subgroup within the group that selects a subset of the
75 bands to be analyzed. The i.group command is also used to prepare this
76 subgroup. The subgroup mechanism allows the user to select a subset of
77 all the band files that form an image.
78 signaturefile=name
80 resultant signature file
81 This is the resultant signature file (containing the means and covari‐
83 ance matrices) for each class in the training map that is associated
84 with the band files in the subgroup selected.
85 maxsig=value
87 maximum number of sub-signatures in any class
88 default: 10
89 The spectral signatures which are produced by this program are "mixed"
91 signatures (see NOTES). Each signature contains one or more subsigna‐
92 tures (represeting subclasses). The algorithm in this program starts
93 with a maximum number of subclasses and reduces this number to a mini‐
94 mal number of subclasses which are spectrally distinct. The user has
95 the option to set this starting value with this option.
96
98 If none of the arguments are specified on the command line, i.gensigset
99 will interactively prompt for the names of these maps and files.
100 It should be noted that interactive mode here only means interactive
102 prompting for maps and files. It does not mean visualization of the
103 signatures that result from the process.
104
106 The algorithm in i.gensigset determines the parameters of a spectral
107 class model known as a Gaussian mixture distribution. The parameters
108 are estimated using multispectral image data and a training map which
109 labels the class of a subset of the image pixels. The mixture class
110 parameters are stored as a class signature which can be used for subse‐
111 quent segmentation (i.e., classification) of the multispectral image.
112 The Gaussian mixture class is a useful model because it can be used to
114 describe the behavior of an information class which contains pixels
115 with a variety of distinct spectral characteristics. For example, for‐
116 est, grasslands or urban areas are examples of information classes that
117 a user may wish to separate in an image. However, each of these infor‐
118 mation classes may contain subclasses each with its own distinctive
119 spectral characteristic. For example, a forest may contain a variety
120 of different tree species each with its own spectral behavior.
121 The objective of mixture classes is to improve segmentation performance
123 by modeling each information class as a probabilistic mixture with a
124 variety of subclasses. The mixture class model also removes the need
125 to perform an initial unsupervised segmentation for the purposes of
126 identifying these subclasses. However, if misclassified samples are
127 used in the training process, these erroneous samples may be grouped as
128 a separate undesired subclass. Therefore, care should be taken to pro‐
129 vided accurate training data.
130 This clustering algorithm estimates both the number of distinct sub‐
132 classes in each class, and the spectral mean and covariance for each
133 subclass. The number of subclasses is estimated using Rissanen's mini‐
134 mum description length (MDL) criteria [1]. This criteria attempts to
135 determine the number of subclasses which "best" describe the data. The
136 approximate maximum likelihood estimates of the mean and covariance of
137 the subclasses are computed using the expectation maximization (EM)
138 algorithm [3].
139
141 J. Rissanen, "A Universal Prior for Integers and Estimation by Minimum
142 Description Length," Annals of Statistics, vol. 11, no. 2, pp. 417-431,
143 1983. A. Dempster, N. Laird and D. Rubin, "Maximum Likelihood from
144 Incomplete Data via the EM Algorithm," J. Roy. Statist. Soc. B, vol.
145 39, no. 1, pp. 1-38, 1977. E. Redner and H. Walker, "Mixture Densi‐
146 ties, Maximum Likelihood and the EM Algorithm," SIAM Review, vol. 26,
147 no. 2, April 1984.
148
150 i.group for creating groups and subgroups
151 v.digit and r.digit for interactively creating the training map.
153 i.smap for creating a final classification layer from the signatures
155 generated by i.gensigset.
156
158 Charles Bouman, School of Electrical Engineering, Purdue University
159 Michael Shapiro, U.S.Army Construction Engineering Research Laboratory
160 Last changed: $Date: 2003-04-17 16:51:33 +0200 (Thu, 17 Apr 2003) $
162 Full index
164 © 2003-2008 GRASS Development Team
166GRASS 6.3.0 i.gensigset(1)