public static InputVector operator +(InputVector v1, InputVector v2)
{
InputVector result = new InputVector();
for (int i = 0; i < SOMConstants.NUM_WEIGHTS; ++i)
{
result.weights[i] = v1.weights[i] + v2.weights[i];
}
return result;
}
public ImageData()
{
m_vectorHistogram = new float[SOMConstants.INPUT_VECTOR_SIZE_HISTOGRAM];
m_vectorArea = new float[SOMConstants.INPUT_VECTOR_SIZE_AREA];
// ccd feature initialization
CEDD = new float[144];
FCTH = new float[192];
JCD = new float[168];
inputVector = new InputVector();
m_BMU.X = -1; m_BMU.Y = -1;
}
private Neuron FindBMU(InputVector inputVec)
{
Neuron winner = null;
float lowestDistance = float.MaxValue;
for (int i = 0; i < SOMConstants.NUM_NODES_DOWN; ++i)
{
for (int j = 0; j < SOMConstants.NUM_NODES_ACROSS; ++j)
{
float dist = SOMHelper.Calculate_Distance(_competitionLayer[i, j].Weights, inputVec.weights);
if (dist < lowestDistance)
{
lowestDistance = dist;
winner = _competitionLayer[i, j];
}
}
}
System.Diagnostics.Debug.Assert(winner != null);
return winner;
}
public Neuron Recognize(InputVector inputVec)
{
Neuron winner = null;
if (IsTrained)
{
float lowestDistance = float.MaxValue;
for (int i = 0; i < SOMConstants.NUM_NODES_DOWN; ++i)
{
for (int j = 0; j < SOMConstants.NUM_NODES_ACROSS; ++j)
{
if (_competitionLayer[i, j].GetImageCount() > 0)
{
float dist = SOMHelper.Calculate_Distance(_competitionLayer[i, j].Weights, inputVec.weights);
if (dist < lowestDistance)
{
lowestDistance = dist;
winner = _competitionLayer[i, j];
}
}
}
}
System.Diagnostics.Debug.Assert(winner != null);
return winner;
}
return winner;
}
public float CalculateDistanceSquared(InputVector inputVec)
{
float distance = 0.0f;
for ( int i = 0; i < inputVec.weights.Length; ++i ) {
distance += (inputVec.weights[ i ] - this._Weights[ i ]) *
(inputVec.weights[ i ] - this._Weights[ i ]);
}
return distance;
}
/// <summary>
/// Calculates the new weight for this node.
/// Equation: W(t+1) = W(t) + THETA(t)*L(t)*(V(t) - W(t))
/// </summary>
/// <param name="targetVector"></param>
/// <param name="learningRate"></param>
/// <param name="influence"></param>
public void AdjustWeights(InputVector targetVector, float learningRate, float influence)
{
for (int w = 0; w < targetVector.weights.Length; ++w)
{
this._Weights[w] += learningRate * influence * (targetVector.weights[w] - this._Weights[w]);
}
}
