BUG: fix hazardous label counting

Modules/Applications/AppClassification/include/otbLearningApplicationBase.h

@@ -111,6 +111,7 @@ public:
   
   typedef typename ModelType::TargetSampleType      TargetSampleType;
   typedef typename ModelType::TargetListSampleType  TargetListSampleType;
+  typedef typename ModelType::TargetValueType       TargetValueType;
   
 #ifdef OTB_USE_OPENCV
   typedef otb::RandomForestsMachineLearningModel<InputValueType, OutputValueType> RandomForestType;

Modules/Applications/AppClassification/include/otbTrainNeuralNetwork.txx

@@ -168,16 +168,14 @@ LearningApplicationBase<TInputValue,TOutputValue>
     }
   else
     {
-    TargetSampleType currentLabel = 0, prevLabel = 0;
+    std::set<TargetValueType> labelSet;
+    TargetSampleType currentLabel;
     for (unsigned int itLab = 0; itLab < trainingLabeledListSample->Size(); ++itLab)
       {
       currentLabel = trainingLabeledListSample->GetMeasurementVector(itLab);
-      if ((currentLabel != prevLabel) || (itLab == 0))
-        {
-        ++nbClasses;
-        }
-      prevLabel = currentLabel;
+      labelSet.insert(currentLabel[0]);
       }
+    nbClasses = labelSet.size();
     layerSizes.push_back(nbClasses);
     }