1i.gensigset(1)                Grass User's Manual               i.gensigset(1)
2
3
4

NAME

6       i.gensigset  - Generates statistics for i.smap from raster map.
7

KEYWORDS

9       imagery, classification, supervised classification, SMAP, signatures
10

SYNOPSIS

12       i.gensigset
13       i.gensigset --help
14       i.gensigset   trainingmap=name   group=name   subgroup=name  signature‐
15       file=name   [maxsig=integer]    [--overwrite]   [--help]    [--verbose]
16       [--quiet]  [--ui]
17
18   Flags:
19       --overwrite
20           Allow output files to overwrite existing files
21
22       --help
23           Print usage summary
24
25       --verbose
26           Verbose module output
27
28       --quiet
29           Quiet module output
30
31       --ui
32           Force launching GUI dialog
33
34   Parameters:
35       trainingmap=name [required]
36           Ground truth training map
37
38       group=name [required]
39           Name of input imagery group
40
41       subgroup=name [required]
42           Name of input imagery subgroup
43
44       signaturefile=name [required]
45           Name for output file containing result signatures
46
47       maxsig=integer
48           Maximum number of sub-signatures in any class
49           Default: 5
50

DESCRIPTION

52       i.gensigset  is  a  non-interactive  method  for  generating input into
53       i.smap.  It is used as the first pass in the a two-pass  classification
54       process.   It  reads a raster map layer, called the training map, which
55       has some of the pixels or regions already classified.  i.gensigset will
56       then extract spectral signatures from an image based on the classifica‐
57       tion of the pixels in the training map and make these signatures avail‐
58       able to i.smap.
59
60       The  user  would  then  execute  the GRASS program i.smap to create the
61       final classified map.
62

OPTIONS

64   Parameters
65       trainingmap=name
66           ground truth training map
67
68       This raster layer, supplied as input by the user, has some of its  pix‐
69       els  already  classified,  and  the  rest (probably most) of the pixels
70       unclassified.  Classified means that the pixel has a non-zero value and
71       unclassified means that the pixel has a zero value.
72
73       This map must be prepared by the user in advance by using a combination
74       of wxGUI vector digitizer and v.to.rast, or some other  import/develop‐
75       ment  process (e.g., v.transects) to define the areas representative of
76       the classes in the image.
77
78       At present, there is no fully-interactive  tool  specifically  designed
79       for producing this layer.
80
81       group=name
82           imagery group
83
84       This  is  the name of the group that contains the band files which com‐
85       prise the image to be analyzed. The i.group command  is  used  to  con‐
86       struct groups of raster layers which comprise an image.
87
88       subgroup=name
89           subgroup containing image files
90
91       This  names  the subgroup within the group that selects a subset of the
92       bands to be analyzed. The i.group command is also used to prepare  this
93       subgroup.  The subgroup mechanism allows the user to select a subset of
94       all the band files that form an image.
95
96       signaturefile=name
97           resultant signature file
98
99       This is the resultant signature file (containing the means and  covari‐
100       ance  matrices)  for  each class in the training map that is associated
101       with the band files in the subgroup selected.
102
103       maxsig=value
104           maximum number of sub-signatures in any class
105           default: 5
106
107       The spectral signatures which are produced by this program are  "mixed"
108       signatures  (see NOTES).  Each signature contains one or more subsigna‐
109       tures (represeting subclasses).  The algorithm in this  program  starts
110       with  a maximum number of subclasses and reduces this number to a mini‐
111       mal number of subclasses which are spectrally distinct.  The  user  has
112       the option to set this starting value with this option.
113

INTERACTIVE MODE

115       If none of the arguments are specified on the command line, i.gensigset
116       will interactively prompt for the names of these maps and files.
117
118       It should be noted that interactive mode here  only  means  interactive
119       prompting  for  maps  and files.  It does not mean visualization of the
120       signatures that result from the process.
121

NOTES

123       The algorithm in i.gensigset determines the parameters  of  a  spectral
124       class  model  known as a Gaussian mixture distribution.  The parameters
125       are estimated using multispectral image data and a training  map  which
126       labels  the  class  of a subset of the image pixels.  The mixture class
127       parameters are stored as a class signature which can be used for subse‐
128       quent segmentation (i.e., classification) of the multispectral image.
129
130       The  Gaussian mixture class is a useful model because it can be used to
131       describe the behavior of an information  class  which  contains  pixels
132       with a variety of distinct spectral characteristics.  For example, for‐
133       est, grasslands or urban areas are examples of information classes that
134       a user may wish to separate in an image.  However, each of these infor‐
135       mation classes may contain subclasses each  with  its  own  distinctive
136       spectral  characteristic.   For example, a forest may contain a variety
137       of different tree species each with its own spectral behavior.
138
139       The objective of mixture classes is to improve segmentation performance
140       by  modeling  each  information class as a probabilistic mixture with a
141       variety of subclasses.  The mixture class model also removes  the  need
142       to  perform  an  initial  unsupervised segmentation for the purposes of
143       identifying these subclasses.  However, if  misclassified  samples  are
144       used in the training process, these erroneous samples may be grouped as
145       a separate undesired subclass.  Therefore, care should be taken to pro‐
146       vided accurate training data.
147
148       This  clustering  algorithm  estimates both the number of distinct sub‐
149       classes in each class, and the spectral mean and  covariance  for  each
150       subclass.  The number of subclasses is estimated using Rissanen’s mini‐
151       mum description length (MDL) criteria [1].  This criteria  attempts  to
152       determine the number of subclasses which "best" describe the data.  The
153       approximate maximum likelihood estimates of the mean and covariance  of
154       the  subclasses  are  computed  using the expectation maximization (EM)
155       algorithm [2,3].
156

WARNINGS

158       If warnings like this occur, reducing the remaining classes to 0:
159       ...
160       WARNING: Removed a singular subsignature number 1 (4 remain)
161       WARNING: Removed a singular subsignature number 1 (3 remain)
162       WARNING: Removed a singular subsignature number 1 (2 remain)
163       WARNING: Removed a singular subsignature number 1 (1 remain)
164       WARNING: Unreliable clustering. Try a smaller initial number of clusters
165       WARNING: Removed a singular subsignature number 1 (-1 remain)
166       WARNING: Unreliable clustering. Try a smaller initial number of clusters
167       Number of subclasses is 0
168       then the user should check for:
169
170           ·   the range of the input data should be between 0 and 100 or  255
171               but  not  between  0.0  and  1.0  (r.info and r.univar show the
172               range)
173
174           ·   the training areas need to contain a sufficient amount of  pix‐
175               els
176

REFERENCES

178           ·   J.  Rissanen, "A Universal Prior for Integers and Estimation by
179               Minimum Description Length," Annals of Statistics, vol. 11, no.
180               2, pp. 417-431, 1983.
181
182           ·   A.  Dempster,  N.  Laird and D. Rubin, "Maximum Likelihood from
183               Incomplete Data via the EM Algorithm," J. Roy. Statist. Soc. B,
184               vol. 39, no. 1, pp. 1-38, 1977.
185
186           ·   E. Redner and H. Walker, "Mixture Densities, Maximum Likelihood
187               and the EM Algorithm," SIAM Review, vol. 26, no. 2, April 1984.
188

SEE ALSO

190        i.group, i.smap, r.info, r.univar, wxGUI vector digitizer
191

AUTHORS

193       Charles Bouman, School of Electrical Engineering, Purdue University
194       Michael Shapiro, U.S.Army Construction Engineering Research Laboratory
195
196       Last changed: $Date: 2018-10-18 21:05:15 +0200 (Thu, 18 Oct 2018) $
197

SOURCE CODE

199       Available at: i.gensigset source code (history)
200
201       Main index | Imagery index | Topics index | Keywords index |  Graphical
202       index | Full index
203
204       © 2003-2019 GRASS Development Team, GRASS GIS 7.6.0 Reference Manual
205
206
207
208GRASS 7.6.0                                                     i.gensigset(1)
Impressum