1i.cluster(1) Grass User's Manual i.cluster(1)
2
6 i.cluster - An imagery function that generates spectral signatures for
7 land cover types in an image using a clustering algorithm. The result‐
8 ing signature file is used as input for i.maxlik, to generate an unsu‐
9 pervised image classification.
10
12 imagery
13
15 i.cluster
16 i.cluster help
17 i.cluster [-q] group=string subgroup=string sigfile=string
18 classes=integer [seed=string] [sample=row_interval,col_interval]
19 [iterations=integer] [convergence=float] [separation=float]
20 [min_size=integer] [reportfile=string]
21 Flags:
23 -q quiet
24 Parameters:
26 group=string
27 Group of imagery files to be clustered
28 subgroup=string
30 Subgroup name in the above group
31 sigfile=string
33 File contains result signatures
34 classes=integer
36 Initial number of classes Options: 1-255
37 seed=string
39 File contains initial signatures
40 sample=row_interval,col_interval
42 Sampling intervals (by row and col); default: ~10,000 pixels
43 iterations=integer
45 Maximum number of iterations Default: 30
46 convergence=float
48 Percent convergence Options: 0-100 Default: 98.0
49 separation=float
51 Cluster separation Default: 0.0
52 min_size=integer
54 Minimum number of pixels in a class Default: 17
55 reportfile=string
57 Name of an output file to contain final report
58
60 i.cluster performs the first pass in the GRASS two-pass unsupervised
61 classification of imagery, while the GRASS program i.maxlik executes
62 the second pass. Both programs must be run to complete the unsupervised
63 classification.
64 i.cluster is a clustering algorithm that reads through the (raster)
66 imagery data and builds pixel clusters based on the spectral
67 reflectances of the pixels. The pixel clusters are imagery categories
68 that can be related to land cover types on the ground. The spectral
69 distributions of the clusters (which will be the land cover spectral
70 signatures) are influenced by six parameters set by the user. The
71 first parameter set by the user is the initial number of clusters to be
72 discriminated. i.cluster starts by generating spectral signatures for
73 this number of clusters and "attempts" to end up with this number of
74 clusters during the clustering process. The resulting number of clus‐
75 ters and their spectral distributions, however, are also influenced by
76 the range of the spectral values (category values) in the image files
77 and the other parameters set by the user. These parameters are: the
78 minimum cluster size, minimum cluster separation, the percent conver‐
79 gence, the maximum number of iterations, and the row and column sam‐
80 pling intervals.
81 The cluster spectral signatures that result are composed of cluster
83 means and covariance matrices. These cluster means and covariance
84 matrices are used in the second pass (i.maxlik) to classify the image.
85 The clusters or spectral classes result can be related to land cover
86 types on the ground. The user has to specify the name of group file,
87 the name of subgroup file, the name of a file to contain result signa‐
88 tures, the initial number of clusters to be discriminated, and option‐
89 ally other parameters (see below) where the group should contain the
90 imagery files that the user wishes to classify. The subgroup is a sub‐
91 set of this group. The user must create a group and subgroup by run‐
92 ning the GRASS program i.group before running i.cluster. The subgroup
93 should contain only the imagery band files that the user wishes to
94 classify. Note that this subgroup must contain more than one band
95 file. The purpose of the group and subgroup is to collect map layers
96 for classification or analysis. The sigfile is the file to contain
97 result signatures which can be used as input for i.maxlik. The classes
98 value is the initial number of clusters to be discriminated; any param‐
99 eter values left unspecified are set to their default values.
100 Flags:
102 -q Run quietly. Suppresses output of program percent-complete mes‐
103 sages and the time elapsed from the beginning of the program. If
104 this flag is not used, these messages are printed out.
105 Parameters:
107 group=name
108 The name of the group file which contains the imagery files that
109 the user wishes to classify.
110 subgroup=name
112 The name of the subset of the group specified in group option,
113 which must contain only imagery band files and more than one
114 band file. The user must create a group and a subgroup by run‐
115 ning the GRASS program i.group before running i.cluster.
116 sigfile=name
118 The name assigned to output signature file which contains signa‐
119 tures of classes and can be used as the input file for the GRASS
120 program i.maxlik for an unsupervised classification.
121 classes=value
123 The number of clusters that will initially be identified in the
124 clustering process before the iterations begin.
125 seed=name
127 The name of a seed signature file is optional. The seed signa‐
128 tures are signatures that contain cluster means and covariance
129 matrices which were calculated prior to the current run of
130 i.cluster. They may be acquired from a previously run of i.clus‐
131 ter or from a supervised classification signature training site
132 section (e.g., using the signature file output by i.class). The
133 purpose of seed signatures is to optimize the cluster decision
134 boundaries (means) for the number of clusters specified.
135 sample=row_interval,col_interval
137 These numbers are optional with default values based on the size
138 of the data set such that the total pixels to be processed is
139 approximately 10,000 (consider round up).
140 iterations=value
142 This parameter determines the maximum number of iterations which
143 is greater than the number of iterations predicted to achieve
144 the optimum percent convergence. The default value is 30. If the
145 number of iterations reaches the maximum designated by the user;
146 the user may want to rerun i.cluster with a higher number of
147 iterations (see reportfile).
148 Default: 30
149 convergence=value
151 A high percent convergence is the point at which cluster means
152 become stable during the iteration process. The default value
153 is 98.0 percent. When clusters are being created, their means
154 constantly change as pixels are assigned to them and the means
155 are recalculated to include the new pixel. After all clusters
156 have been created, i.cluster begins iterations that change clus‐
157 ter means by maximizing the distances between them. As these
158 means shift, a higher and higher convergence is approached.
159 Because means will never become totally static, a percent con‐
160 vergence and a maximum number of iterations are supplied to stop
161 the iterative process. The percent convergence should be
162 reached before the maximum number of iterations. If the maximum
163 number of iterations is reached, it is probable that the desired
164 percent convergence was not reached. The number of iterations is
165 reported in the cluster statistics in the report file (see
166 reportfile).
167 Default: 98.0
168 separation=value
170 This is the minimum separation below which clusters will be
171 merged in the iteration process. The default value is 0.0. This
172 is an image-specific number (a "magic" number) that depends on
173 the image data being classified and the number of final clusters
174 that are acceptable. Its determination requires experimentation.
175 Note that as the minimum class (or cluster) separation is
176 increased, the maximum number of iterations should also be
177 increased to achieve this separation with a high percentage of
178 convergence (see convergence).
179 Default: 0.0
180 min_size=value
182 This is the minimum number of pixels that will be used to define
183 a cluster, and is therefore the minimum number of pixels for
184 which means and covariance matrices will be calculated.
185 Default: 17
186 reportfile=name
188 The reportfile is an optional parameter which contains the
189 result, i.e., the statistics for each cluster. Also included are
190 the resulting percent convergence for the clusters, the number
191 of iterations that was required to achieve the convergence, and
192 the separability matrix.
193
195 Running in command line mode, i.cluster will overwrite the output sig‐
196 nature file and reportfile (if required by the user) without prompting
197 if the files existed.
198
200 GRASS Tutorial: Image Processing
201 i.class
203 i.group
204 i.gensig
205 i.maxlik
206
208 Michael Shapiro, U.S.Army Construction Engineering Research Laboratory
209 Tao Wen, University of Illinois at Urbana-Champaign, Illinois
210 Last changed: $Date: 2003/04/17 14:48:43 $
212 Full index
214GRASS 6.2.2 i.cluster(1)