public ClassificationType[,] Classify(LasFile file)
{
var sw = Stopwatch.StartNew();
ClassificationType[,] output = new ClassificationType[_divisionCountX, _divisionCountY];
SubgroupOfPoints[,] values = Utills.GroupPoints(file, _divisionCountX, _divisionCountY);
Statistics stats = new Statistics();
stats.Count = _divisionCountX * _divisionCountY;
for (int i = 0; i < _divisionCountX; i++)
{
for (int j = 0; j < _divisionCountY; j++)
{
double avgHeight = values[i, j].avgHeight;
double avgIntensity = values[i, j].avgIntensity;
double avgDistance = values[i, j].avgDistance;
OpenTK.Vector3 slopeVector = values[i, j].slopeVector;
IMLData classed = Network.Compute(new BasicMLData(
new double[] { avgDistance, avgHeight, avgIntensity, slopeVector[0], slopeVector[1], slopeVector[2] }));
output[i, j] = Utills.QuickClassess[classed.IndexOfMax()];
ClassificationType ct;
if (!Utills.QuickClassess.TryGetValue(values[i, j].classIndex, out ct))
{
continue;
}
if (output[i, j] != ct)
{
stats.ClassErrors[(int)ct]++;
}
stats.PredictionMatrix[(int)output[i, j], (int)ct]++;
stats.PredictionMatrix[(int)ct, (int)output[i, j]]++;
stats.ClassCount[(int)output[i, j]]++;
stats.ClassRealCount[(int)ct]++;
}
Console.WriteLine(i);
}
Console.Write(stats.ToString());
sw.Stop();
Console.WriteLine("Czas trwania [" + sw.Elapsed.TotalSeconds.ToString() + "s]");
stats.SaveMatrixAsCSV();
return(output);
}
public static Tuple <double[], int> GetInputOutputFromClassificationType(LasPoint point, LasFile file)
{
var abc = LinearRegression.ComputeRegressionPoint(file, point, regressionCount, regressionRange);
double distanceFromPlane = Utills.DistanceFromPlane(point, abc);
LasPoint3Short pointShort = (LasPoint3Short)point;
double green = pointShort.Green - (pointShort.Red + pointShort.Blue) / 2;
var input = new double[] { green, file.LasHeader.ScaleZ(point.Z), point.Intensity, abc.X, abc.Y, abc.Z, distanceFromPlane };
int output;
switch (point.Classification)
{
case LasPoint.ClassificationType.Ground:
output = 0;
break;
case LasPoint.ClassificationType.HighVegetation:
output = 1;
break;
case LasPoint.ClassificationType.Building:
output = 2;
break;
case LasPoint.ClassificationType.MediumVegetation:
output = 3;
break;
case LasPoint.ClassificationType.LowVegetation:
output = 4;
break;
case LasPoint.ClassificationType.Water:
output = 5;
break;
default:
input = new double[] { -1, -1, -1, -1, -1, -1, -1 };
output = 6;
break;
}
return(Tuple.Create(input, output));
}
public EncogNeuralNetworkQuick(LasFile file, int divisionCountX, int divisionCountY)
{
var sw = Stopwatch.StartNew();
_divisionCountX = divisionCountX;
_divisionCountY = divisionCountY;
var groupPointList = Utills.GroupPointsList(file, _divisionCountX, _divisionCountY);
int count = groupPointList.Count / 5;
double[][] input = new double[count][];
double[][] ideal = new double[count][];
int waterCount = 0;
int groundCount = 0;
int lowCount = 0;
int mediumCount = 0;
int highCount = 0;
int buildingCount = 0;
for (int i = 0; i < count; i++)
{
int rndNumber;
ClassificationType simpleClass;
while (true)
{
rndNumber = _rnd.Next(0, groupPointList.Count - 1);
if (!Utills.QuickClassess.TryGetValue(groupPointList[rndNumber].classIndex, out simpleClass))
{
continue;
}
if (simpleClass == LasPoint.ClassificationType.Water)
{
waterCount++;
if (waterCount - 25 < count / 6)
{
break;
}
}
else if (simpleClass == LasPoint.ClassificationType.Ground)
{
groundCount++;
if (groundCount - 25 < count / 6)
{
break;
}
}
else if (simpleClass == LasPoint.ClassificationType.Building)
{
buildingCount++;
if (buildingCount - 25 < count / 6)
{
break;
}
}
else if (simpleClass == LasPoint.ClassificationType.LowVegetation)
{
lowCount++;
if (lowCount - 25 < count / 6)
{
break;
}
}
else if (simpleClass == LasPoint.ClassificationType.MediumVegetation)
{
mediumCount++;
if (mediumCount - 25 < count / 6)
{
break;
}
}
else if (simpleClass == LasPoint.ClassificationType.HighVegetation)
{
highCount++;
if (highCount - 25 < count / 6)
{
break;
}
}
if (highCount > 5 * count)
{
highCount = 0;
}
if (buildingCount > 5 * count)
{
buildingCount = 0;
}
if (lowCount > 5 * count)
{
lowCount = 0;
}
if (mediumCount > 5 * count)
{
mediumCount = 0;
}
if (waterCount > 5 * count)
{
waterCount = 0;
}
if (groundCount > 5 * count)
{
groundCount = 0;
}
}
if (i % 100 == 0)
{
Console.WriteLine(i);
}
double avgHeight = groupPointList[rndNumber].avgHeight;
double avgIntensity = groupPointList[rndNumber].avgIntensity;
double avgDistance = groupPointList[rndNumber].avgDistance;
OpenTK.Vector3 slopeVector = groupPointList[rndNumber].slopeVector;
input[i] = new double[] { avgDistance, avgHeight, avgIntensity, slopeVector[0], slopeVector[1], slopeVector[2] };
ideal[i] = Utills.ClassToVector(simpleClass);
}
inputNumber = input[0].Length;
init();
IMLDataSet trainingSet = new BasicMLDataSet(input, ideal);
IMLTrain train = new ResilientPropagation(Network, trainingSet);
int epoch = 1;
do
{
train.Iteration();
Console.WriteLine(train.Error + " | " + epoch);
epoch++;
} while (epoch < 1000);
LearningError = train.Error;
train.FinishTraining();
sw.Stop();
Console.WriteLine("Czas trwania [" + sw.Elapsed.TotalSeconds.ToString() + "s]");
}
public EncogNeuralNetworkSlow(LasFile file)
{
var sw = Stopwatch.StartNew();
int count = 300000;
LasPointDataRecords points = file.LasPointDataRecords;
double[][] input = new double[count][];
double[][] ideal = new double[count][];
int waterCount = 0;
int groundCount = 0;
int lowCount = 0;
int mediumCount = 0;
int highCount = 0;
int buildingCount = 0;
for (int i = 0; i < count; i++)
{
int rndNumber;
while (true)
{
rndNumber = _rnd.Next(0, points.Count - 1);
if (points[rndNumber].Classification == LasPoint.ClassificationType.Water)
{
waterCount++;
if (waterCount - 25 < count / 6)
{
break;
}
}
else if (points[rndNumber].Classification == LasPoint.ClassificationType.Ground)
{
groundCount++;
if (groundCount - 25 < count / 6)
{
break;
}
}
else if (points[rndNumber].Classification == LasPoint.ClassificationType.Building)
{
buildingCount++;
if (buildingCount - 25 < count / 6)
{
break;
}
}
else if (points[rndNumber].Classification == LasPoint.ClassificationType.LowVegetation)
{
lowCount++;
if (lowCount - 25 < count / 6)
{
break;
}
}
else if (points[rndNumber].Classification == LasPoint.ClassificationType.MediumVegetation)
{
mediumCount++;
if (mediumCount - 25 < count / 6)
{
break;
}
}
else if (points[rndNumber].Classification == LasPoint.ClassificationType.HighVegetation)
{
highCount++;
if (highCount - 25 < count / 6)
{
break;
}
}
if (highCount > 5 * count)
{
highCount = 0;
}
if (buildingCount > 5 * count)
{
buildingCount = 0;
}
if (lowCount > 5 * count)
{
lowCount = 0;
}
if (mediumCount > 5 * count)
{
mediumCount = 0;
}
if (waterCount > 5 * count)
{
waterCount = 0;
}
if (groundCount > 5 * count)
{
groundCount = 0;
}
}
if (i % 1000 == 0)
{
Console.WriteLine("Selected point: " + i + "/" + count);
}
//double[] regression = LinearRegression.ComputeRegressionNumerics(file, points[rndNumber], regressionCount, regressionRange);
OpenTK.Vector3 abc = LinearRegression.ComputeRegressionPoint(file, points[rndNumber], regressionCount, regressionRange);
LasPoint3Short point = (LasPoint3Short)points[rndNumber];
double distanceFromPlane = Utills.DistanceFromPlane(point, abc);
double green = point.Green - (point.Red + point.Blue) / 2;
input[i] = new double[] { green, file.LasHeader.ScaleZ(point.Z), point.Intensity, abc.X, abc.Y, abc.Z, distanceFromPlane };
ideal[i] = Utills.ClassToVector(point.Classification);
}
inputNumber = input[0].Length;
init();
IMLDataSet trainingSet = new BasicMLDataSet(input, ideal);
IMLTrain train = new ResilientPropagation(Network, trainingSet);
int epoch = 1;
do
{
train.Iteration();
Console.WriteLine("Train error: " + train.Error + ", iteration: " + epoch);
epoch++;
} while (epoch < 1000);
LearningError = train.Error;
train.FinishTraining();
sw.Stop();
Console.WriteLine("Czas trwania [" + sw.Elapsed.TotalSeconds.ToString() + "s]");
}
public EncogNeuralNetwork(LasFile file)
{
Stopwatch sw = Stopwatch.StartNew();
int count = 300000;
LasPointDataRecords points = file.LasPointDataRecords;
double[][] input = new double[count][];
double[][] ideal = new double[count][];
int waterCount = 0;
int groundCount = 0;
int lowCount = 0;
int mediumCount = 0;
int highCount = 0;
int buildingCount = 0;
for (int i = 0; i < count; i++)
{
int rndNumber;
while (true)
{
rndNumber = _rnd.Next(0, points.Count - 1);
if (points[rndNumber].Classification == LasPoint.ClassificationType.Water)
{
waterCount++;
if (waterCount - 25 < count / 6)
{
break;
}
}
else if (points[rndNumber].Classification == LasPoint.ClassificationType.Ground)
{
groundCount++;
if (groundCount - 25 < count / 6)
{
break;
}
}
else if (points[rndNumber].Classification == LasPoint.ClassificationType.Building)
{
buildingCount++;
if (buildingCount - 25 < count / 6)
{
break;
}
}
else if (points[rndNumber].Classification == LasPoint.ClassificationType.LowVegetation)
{
lowCount++;
if (lowCount - 25 < count / 6)
{
break;
}
}
else if (points[rndNumber].Classification == LasPoint.ClassificationType.MediumVegetation)
{
mediumCount++;
if (mediumCount - 25 < count / 6)
{
break;
}
}
else if (points[rndNumber].Classification == LasPoint.ClassificationType.HighVegetation)
{
highCount++;
if (highCount - 25 < count / 6)
{
break;
}
}
if (highCount > 5 * count)
{
highCount = 0;
}
if (buildingCount > 5 * count)
{
buildingCount = 0;
}
if (lowCount > 5 * count)
{
lowCount = 0;
}
if (mediumCount > 5 * count)
{
mediumCount = 0;
}
if (waterCount > 5 * count)
{
waterCount = 0;
}
if (groundCount > 5 * count)
{
groundCount = 0;
}
}
if (i % 100 == 0)
{
Console.WriteLine(i);
}
LasPoint3Short point = (LasPoint3Short)points[rndNumber];
double green = point.Green - (point.Red + point.Blue) / 2;
input[i] = new double[] { file.LasHeader.ScaleZ(point.Z), point.Intensity, green };
ideal[i] = Utills.ClassToVector(point.Classification);
}
inputNumber = input[0].Length;
init();
IMLDataSet trainingSet = new BasicMLDataSet(input, ideal);
IMLTrain train = new ResilientPropagation(Network, trainingSet);
int epoch = 1;
do
{
train.Iteration();
Console.WriteLine(train.Error + " | " + epoch);
epoch++;
} while (epoch < 1000);
LearningError = train.Error;
train.FinishTraining();
sw.Stop();
Console.WriteLine("Czas trwania [" + sw.Elapsed.TotalSeconds.ToString() + "s]");
}
public ClassificationType[,] Classify(LasFile file, int divCountX, int divCountY)
{
Stopwatch swTotal = Stopwatch.StartNew();
Stopwatch sw = Stopwatch.StartNew();
Console.WriteLine("Preparing testing dataset...");
LasPointDataRecords points = file.LasPointDataRecords;
ClassificationType[,] output = new ClassificationType[divCountX, divCountY];
SubgroupOfPoints[,] values = Utills.GroupPoints(file, divCountX, divCountY);
Statistics stats = new Statistics();
stats.Count = divCountX * divCountY;
sw.Stop();
Console.WriteLine("Preparing testing dataset completed [" + sw.Elapsed.TotalSeconds.ToString() + "s]");
Stopwatch sw2 = Stopwatch.StartNew();
Console.WriteLine("Classification in progress...");
int noiseCount = 0;
for (int i = 0; i < divCountX; i++)
{
for (int j = 0; j < divCountY; j++)
{
if (values[i, j].classIndex == 7)
{
output[i, j] = ClassificationType.Noise;
noiseCount++;
}
else
{
double avgHeight = values[i, j].avgHeight;
double avgIntensity = values[i, j].avgIntensity;
double avgDistance = values[i, j].avgDistance;
OpenTK.Vector3 slopeVector = values[i, j].slopeVector;
output[i, j] = Utills.ClassificationClasses[knn.Compute(new double[] {
avgDistance, avgHeight, avgIntensity, slopeVector[0],
slopeVector[1], slopeVector[2]
})];
ClassificationType ct;
if (!Utills.QuickClassess.TryGetValue(values[i, j].classIndex, out ct))
{
continue;
}
if (output[i, j] != ct)
{
stats.ClassErrors[(int)ct]++;
}
stats.PredictionMatrix[(int)output[i, j], (int)ct]++;
stats.PredictionMatrix[(int)ct, (int)output[i, j]]++;
stats.ClassCount[(int)output[i, j]]++;
stats.ClassRealCount[(int)ct]++;
}
}
//Console.WriteLine(i);
}
Console.Write(stats.ToString());
sw2.Stop();
Console.WriteLine("Classification completed [" + sw2.Elapsed.TotalSeconds.ToString() + "s]");
swTotal.Stop();
Console.WriteLine("Total time: [" + swTotal.Elapsed.TotalSeconds.ToString() + "s]");
Console.WriteLine("Noise count: " + noiseCount.ToString());
stats.SaveMatrixAsCSV();
return(output);
}
