i.gensigset: fix possible pole and divide by zero errors in regroup

Using logarithm function call with zero argument will lead to
a pole error, which occurs if the mathematical function has
an exact infinite result.

Check if the argument value is zero before passing that to
the log function to avoid such errors. I also added check
for negative numbers just to make sure the argument is in
the right domain for the log function as well.

There was also a possible divide by zero scenario when we
were dividing the class data by subsum, which can be zero.
Added a conditional check which avoids going to that stage.

Signed-off-by: Mohan Yelugoti <[email protected]>

imagery/i.gensigset/subcluster.c

@@ -366,10 +366,14 @@ static double regroup(struct ClassSig *Sig, int nbands)
             subsum += tmp;
             Data->p[s][i] = tmp;
         }
-        likelihood += log(subsum) + maxlike;
 
-        for (i = 0; i < Sig->nsubclasses; i++)
-            Data->p[s][i] /= subsum;
+        if (subsum > 0)
+            likelihood += log(subsum) + maxlike;
+
+        if (subsum) {
+            for (i = 0; i < Sig->nsubclasses; i++)
+                Data->p[s][i] /= subsum;
+        }
     }
 
     return (likelihood);