<div>NO NO NO!</div>
<div> </div>
<div>I have 30 CORINE land cover classes to classify.</div>
<div>The number 512, for example, means 5.1.2, is a </div>
<div>nominal number, IS NOT an classe id.</div>
<div> </div>
<div>But may be right: may be the way the classes are </div>
<div>represented in the system cause that problem. I&#39;ll test</div>
<div>that..</div>
<div> </div>
<div> </div>
<div>Thanks</div>
<div>JD<br><br></div>
<div class="gmail_quote">On Tue, Apr 6, 2010 at 6:44 PM, Micha Silver <span dir="ltr">&lt;<a href="mailto:micha@arava.co.il">micha@arava.co.il</a>&gt;</span> wrote:<br>
<blockquote class="gmail_quote" style="PADDING-LEFT: 1ex; MARGIN: 0px 0px 0px 0.8ex; BORDER-LEFT: #ccc 1px solid">joel dinis wrote:<br><br>
<blockquote class="gmail_quote" style="PADDING-LEFT: 1ex; MARGIN: 0px 0px 0px 0.8ex; BORDER-LEFT: #ccc 1px solid">Hello Micha,<br><br>I put <a href="http://r.info/" target="_blank">r.info</a> &lt;<a href="http://r.info/" target="_blank">http://r.info/</a>&gt; -r amostras 
<div class="im"><br>as sugested, get the following:<br>min=112<br>max=512<br><br></div></blockquote>
<div class="im">So, if I understand correctly, you are trying to classify into 400 separate classes?!<br>I&#39;m a beginner with image classification, and others with more experience will know better, but I&#39;ll bet that&#39;s your problem. There must be many separate training areas (with different cat values) with a very similar spectral signature, and the algorithm is not able to differentiate between them.<br>
<br></div>
<blockquote class="gmail_quote" style="PADDING-LEFT: 1ex; MARGIN: 0px 0px 0px 0.8ex; BORDER-LEFT: #ccc 1px solid">
<div class="im">(This values are index to my classes.)<br><br>JD<br><br></div>
<div class="im">On Tue, Apr 6, 2010 at 4:39 PM, Micha Silver &lt;<a href="mailto:micha@arava.co.il" target="_blank">micha@arava.co.il</a> &lt;mailto:<a href="mailto:micha@arava.co.il" target="_blank">micha@arava.co.il</a>&gt;&gt; wrote:<br>
<br>   Nikos Alexandris wrote:<br><br>       (sorry for re-sorting the message, just to keep it easy to<br>       follow-up)<br><br>       joel dinis:<br>        <br>                  &gt; I trying to classify remote sensing imagery, and<br>
                  &gt; I thought to try the i.smap module, but<br>                  &gt; when I run the i.gensigset,<br>              <br><br>       Nikos Alexandris:<br>        <br>                  Please, could you copy-paste the exact command you<br>
           used?<br>                  Does the &quot;maxsig=&quot; (e.g. maxsig=20) make any<br>           difference?<br>              <br><br>       joel dinis:<br><br>        <br>           In fact, I put maxsig=30.<br>
           i.gensigset trainingmap=amostras group=grupo subgroup=subgrupo<br>           signaturefile=assinaturas maxsig=30<br>              <br><br>       What about smaller maxsig&#39;s, like:<br><br>       - maxsig=25<br>
       - maxsig=20<br>       - maxsig=15<br><br>       ?<br><br>       Nikos<br><br>        <br>   Also it might help to know what is contained in the trainingmap. i.e:<br></div>   <a href="http://r.info/" target="_blank">r.info</a> &lt;<a href="http://r.info/" target="_blank">http://r.info</a>&gt; -r amostras 
<div>
<div></div>
<div class="h5"><br><br>        <br>                          &gt;  I always<br>                  &gt; get the same problem:<br>                  &gt; ...<br>                  &gt; Removed a singular subsignature number 1 (9 remain)<br>
                  &gt; Removed a singular subsignature number 1 (8 remain)<br>                  &gt; Removed a singular subsignature number 1 (7 remain)<br>                  &gt; Removed a singular subsignature number 1 (6 remain)<br>
                  &gt; Removed a singular subsignature number 1 (5 remain)<br>                  &gt; Removed a singular subsignature number 1 (4 remain)<br>                  &gt; Removed a singular subsignature number 1 (3 remain)<br>
                  &gt; Removed a singular subsignature number 1 (2 remain)<br>                  &gt; Removed a singular subsignature number 1 (1 remain)<br>                  &gt; Unreliable clustering. Try a smaller initial<br>
           number of<br>                  clusters<br>                  &gt; Removed a singular subsignature number 1 (-1 remain)<br>                  &gt; Unreliable clustering. Try a smaller initial<br>           number of<br>
                  clusters<br>                  &gt; Number of subclasses is 0<br>                  &gt; Clustering class 12 (184 pixels)...<br>                  &gt; Removed a singular subsignature number 1 (9 remain)<br>
                  &gt; Removed a singular subsignature number 1 (8 remain)<br>                  &gt; Removed a singular subsignature number 1 (7 remain)<br>                  &gt; Removed a singular subsignature number 1 (6 remain)<br>
                  &gt; Removed a singular subsignature number 1 (5 remain)<br>                  &gt; Removed a singular subsignature number 1 (4 remain)<br>                  &gt; Removed a singular subsignature number 1 (3 remain)<br>
                  &gt; Removed a singular subsignature number 1 (2 remain)<br>                  &gt; Removed a singular subsignature number 1 (1 remain)<br>                  &gt; Unreliable clustering. Try a smaller initial<br>
           number of<br>                  clusters<br>                  &gt; Removed a singular subsignature number 1 (-1 remain)<br>                  &gt; Unreliable clustering. Try a smaller initial<br>           number of<br>
                  clusters<br>                  &gt; Number of subclasses is 0<br>                  &gt; i.gensigset complete.<br>                  &gt;<br>                  &gt; And afterwards the signature file doesn&#39;t have any<br>
                  &gt; spectral information. I really don&#39;t understand whats<br>                  happening<br>                  &gt; and the cause to this outcome.<br>                  &gt;<br>                  &gt; Can anyone solve this &quot;enigma&quot;?<br>
                  &gt;<br>                  &gt; Ps: I don&#39;t know if this information is important: I<br>                  collected<br>                  &gt; about 450 samples distributed by 10 classes. I<br>           may also<br>
                  &gt; add that the average number of pixels per sample<br>           is around<br>                  50.<br>                  &gt;<br>                  &gt; And thanks list, for the previous tips: they were<br>
           very<br>                  useful.<br>              <br><br><br><br>       _______________________________________________<br>       grass-user mailing list<br></div></div>       <a href="mailto:grass-user@lists.osgeo.org" target="_blank">grass-user@lists.osgeo.org</a> &lt;mailto:<a href="mailto:grass-user@lists.osgeo.org" target="_blank">grass-user@lists.osgeo.org</a>&gt; 
<div class="im"><br>       <a href="http://lists.osgeo.org/mailman/listinfo/grass-user" target="_blank">http://lists.osgeo.org/mailman/listinfo/grass-user</a><br><br>       This mail was received via Mail-SeCure System.<br>
<br><br>        <br><br><br>   --     Micha Silver<br>   Arava Development Co. +972-52-3665918<br>   <a href="http://www.surfaces.co.il/" target="_blank">http://www.surfaces.co.il</a><br><br><br><br><br>This mail was received via Mail-SeCure System.<br>
</div></blockquote>
<div>
<div></div>
<div class="h5"><br><br>-- <br>Micha Silver<br>Arava Development Co. +972-52-3665918<br><a href="http://www.surfaces.co.il/" target="_blank">http://www.surfaces.co.il</a><br><br><br></div></div></blockquote></div><br>