otbVectorClassifier.cxx 14.7 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91
/*
 * Copyright (C) 2005-2017 Centre National d'Etudes Spatiales (CNES)
 *
 * This file is part of Orfeo Toolbox
 *
 *     https://www.orfeo-toolbox.org/
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

#include "otbWrapperApplication.h"
#include "otbWrapperApplicationFactory.h"

#include "otbOGRDataSourceWrapper.h"
#include "otbOGRFeatureWrapper.h"

#include "itkVariableLengthVector.h"
#include "otbStatisticsXMLFileReader.h"

#include "itkListSample.h"
#include "otbShiftScaleSampleListFilter.h"

#include "otbMachineLearningModelFactory.h"

#include "otbMachineLearningModel.h"

#include <time.h>

namespace otb
{
namespace Wrapper
{

/** Utility function to negate std::isalnum */
bool IsNotAlphaNum(char c)
  {
  return !std::isalnum(c);
  }

class VectorClassifier : public Application
{
public:
  /** Standard class typedefs. */
  typedef VectorClassifier              Self;
  typedef Application                   Superclass;
  typedef itk::SmartPointer<Self>       Pointer;
  typedef itk::SmartPointer<const Self> ConstPointer;

  /** Standard macro */
  itkNewMacro(Self);

  itkTypeMacro(Self, Application)

  /** Filters typedef */
  typedef double                                        ValueType;
  typedef unsigned int                                  LabelType;
  typedef itk::FixedArray<LabelType,1>                  LabelSampleType;
  typedef itk::Statistics::ListSample<LabelSampleType>  LabelListSampleType;

  typedef otb::MachineLearningModel<ValueType,LabelType>          MachineLearningModelType;
  typedef otb::MachineLearningModelFactory<ValueType, LabelType>  MachineLearningModelFactoryType;
  typedef MachineLearningModelType::Pointer                       ModelPointerType;
  typedef MachineLearningModelType::ConfidenceListSampleType      ConfidenceListSampleType;

  /** Statistics Filters typedef */
  typedef itk::VariableLengthVector<ValueType>                    MeasurementType;
  typedef otb::StatisticsXMLFileReader<MeasurementType>           StatisticsReader;

  typedef itk::VariableLengthVector<ValueType>                    InputSampleType;
  typedef itk::Statistics::ListSample<InputSampleType>            ListSampleType;
  typedef otb::Statistics::ShiftScaleSampleListFilter<ListSampleType, ListSampleType> ShiftScaleFilterType;

  ~VectorClassifier() ITK_OVERRIDE
    {
    MachineLearningModelFactoryType::CleanFactories();
    }

private:
  void DoInit() ITK_OVERRIDE
  {
    SetName("VectorClassifier");
92 93
    SetDescription("Performs a classification of the input vector data according to a model file."
      "Features of the vector data output will contain the class labels decided by the classifier (maximal class label = 65535).");
94 95 96 97

    SetDocName("Vector Classification");
    SetDocAuthors("OTB-Team");
    SetDocLongDescription("This application performs a vector data classification based on a model file produced by the TrainVectorClassifier application.");
98
    SetDocLimitations("Shapefiles are supported. But the SQLite format is only supported in update mode.");
99
    SetDocSeeAlso("TrainVectorClassifier");
100
    AddDocTag(Tags::Learning);
101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147

    AddParameter(ParameterType_InputVectorData, "in", "Name of the input vector data");
    SetParameterDescription("in","The input vector data to classify.");

    AddParameter(ParameterType_InputFilename, "instat", "Statistics file");
    SetParameterDescription("instat", "A XML file containing mean and standard deviation to center"
      "and reduce samples before classification (produced by ComputeImagesStatistics application).");
    MandatoryOff("instat");

    AddParameter(ParameterType_InputFilename, "model", "Model file");
    SetParameterDescription("model", "A model file (produced by TrainVectorClassifier application,"
      "maximal class label = 65535).");

    AddParameter(ParameterType_String,"cfield","Field containing the predicted class");
    SetParameterDescription("cfield","Field containing the predicted class."
      "Only geometries with this field available will be taken into account.");
    SetParameterString("cfield","predicted", false);

    AddParameter(ParameterType_ListView, "feat", "Field names to be calculated."); //
    SetParameterDescription("feat","List of field names in the input vector data used as features for training."); //

    AddParameter(ParameterType_Empty, "confmap",  "Confidence map");
    SetParameterDescription( "confmap", "Confidence map of the produced classification. The confidence index depends on the model : \n"
      "  - LibSVM : difference between the two highest probabilities (needs a model with probability estimates, so that classes probabilities can be computed for each sample)\n"
      "  - OpenCV\n"
      "    * Boost : sum of votes\n"
      "    * DecisionTree : (not supported)\n"
      "    * GradientBoostedTree : (not supported)\n"
      "    * KNearestNeighbors : number of neighbors with the same label\n"
      "    * NeuralNetwork : difference between the two highest responses\n"
      "    * NormalBayes : (not supported)\n"
      "    * RandomForest : Confidence (proportion of votes for the majority class). Margin (normalized difference of the votes of the 2 majority classes) is not available for now.\n"
      "    * SVM : distance to margin (only works for 2-class models)\n");
    MandatoryOff("confmap");

    AddParameter(ParameterType_OutputFilename, "out", "Output vector data file containing class labels");
    SetParameterDescription("out","Output vector data file storing sample values (OGR format)."
      "If not given, the input vector data file is updated.");
    MandatoryOff("out");

    // Doc example parameter settings
    SetDocExampleParameterValue("in", "vectorData.shp");
    SetDocExampleParameterValue("instat", "meanVar.xml");
    SetDocExampleParameterValue("model", "svmModel.svm");
    SetDocExampleParameterValue("out", "svmModel.svm");
    SetDocExampleParameterValue("feat", "perimeter  area  width");
    SetDocExampleParameterValue("cfield", "predicted");
148

149
    SetOfficialDocLink();
150 151 152 153 154 155 156 157 158 159 160 161 162 163
  }

  void DoUpdateParameters() ITK_OVERRIDE
  {
    if ( HasValue("in") )
    {
      std::string shapefile = GetParameterString("in");

      otb::ogr::DataSource::Pointer ogrDS;

      OGRSpatialReference oSRS("");
      std::vector<std::string> options;

      ogrDS = otb::ogr::DataSource::New(shapefile, otb::ogr::DataSource::Modes::Read);
164 165
      otb::ogr::Layer layer = ogrDS->GetLayer(0);
      OGRFeatureDefn &layerDefn = layer.GetLayerDefn();
166 167 168

      ClearChoices("feat");

169
      for(int iField=0; iField< layerDefn.GetFieldCount(); iField++)
170
      {
171 172 173 174
        std::string item = layerDefn.GetFieldDefn(iField)->GetNameRef();
        std::string key(item);
        key.erase( std::remove_if(key.begin(),key.end(),IsNotAlphaNum), key.end());
        std::transform(key.begin(), key.end(), key.begin(), tolower);
175

176
        OGRFieldType fieldType = layerDefn.GetFieldDefn(iField)->GetType();
177 178
        if(fieldType == OFTInteger ||  ogr::version_proxy::IsOFTInteger64(fieldType) || fieldType == OFTReal)
          {
179
          std::string tmpKey="feat."+key;
180 181 182 183 184 185 186 187 188 189
          AddChoice(tmpKey,item);
          }
      }
    }
  }

  void DoExecute() ITK_OVERRIDE
  {
    clock_t tic = clock();

190
    std::string shapefile = GetParameterString("in");
191 192 193 194 195 196 197 198 199

    otb::ogr::DataSource::Pointer source = otb::ogr::DataSource::New(shapefile, otb::ogr::DataSource::Modes::Read);
    otb::ogr::Layer layer = source->GetLayer(0);

    ListSampleType::Pointer input = ListSampleType::New();

    const int nbFeatures = GetSelectedItems("feat").size();
    input->SetMeasurementVectorSize(nbFeatures);

200 201 202
    otb::ogr::Layer::const_iterator it = layer.cbegin();
    otb::ogr::Layer::const_iterator itEnd = layer.cend();
    for( ; it!=itEnd ; ++it)
203
      {
204 205 206
      MeasurementType mv;
      mv.SetSize(nbFeatures);
      for(int idx=0; idx < nbFeatures; ++idx)
207
        {
208
        mv[idx] = (*it)[GetSelectedItems("feat")[idx]].GetValue<double>();
209
        }
210
      input->PushBack(mv);
211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251
      }

    // Statistics for shift/scale
    MeasurementType meanMeasurementVector;
    MeasurementType stddevMeasurementVector;
    if (HasValue("instat") && IsParameterEnabled("instat"))
      {
      StatisticsReader::Pointer statisticsReader = StatisticsReader::New();
      std::string XMLfile = GetParameterString("instat");
      statisticsReader->SetFileName(XMLfile);
      meanMeasurementVector = statisticsReader->GetStatisticVectorByName("mean");
      stddevMeasurementVector = statisticsReader->GetStatisticVectorByName("stddev");
      }
    else
      {
      meanMeasurementVector.SetSize(nbFeatures);
      meanMeasurementVector.Fill(0.);
      stddevMeasurementVector.SetSize(nbFeatures);
      stddevMeasurementVector.Fill(1.);
      }

    ShiftScaleFilterType::Pointer trainingShiftScaleFilter = ShiftScaleFilterType::New();
    trainingShiftScaleFilter->SetInput(input);
    trainingShiftScaleFilter->SetShifts(meanMeasurementVector);
    trainingShiftScaleFilter->SetScales(stddevMeasurementVector);
    trainingShiftScaleFilter->Update();
    otbAppLogINFO("mean used: " << meanMeasurementVector);
    otbAppLogINFO("standard deviation used: " << stddevMeasurementVector);

    otbAppLogINFO("Loading model");
    m_Model = MachineLearningModelFactoryType::CreateMachineLearningModel(GetParameterString("model"),
                                                MachineLearningModelFactoryType::ReadMode);

    if (m_Model.IsNull())
      {
      otbAppLogFATAL(<< "Error when loading model " << GetParameterString("model") << " : unsupported model type");
      }

    m_Model->Load(GetParameterString("model"));
    otbAppLogINFO("Model loaded");

252
    ListSampleType::Pointer listSample = trainingShiftScaleFilter->GetOutput();
253

254
    ConfidenceListSampleType::Pointer quality;
255 256 257 258 259 260 261 262 263

    bool computeConfidenceMap(IsParameterEnabled("confmap") && m_Model->HasConfidenceIndex() 
                              && !m_Model->GetRegressionMode());

    if (!m_Model->HasConfidenceIndex() && IsParameterEnabled("confmap"))
      {
      otbAppLogWARNING("Confidence map requested but the classifier doesn't support it!");
      }

264
    LabelListSampleType::Pointer target;
265 266 267 268 269 270 271 272 273 274 275
    if (computeConfidenceMap)
      {
      quality = ConfidenceListSampleType::New();
      target = m_Model->PredictBatch(listSample, quality);
      }
      else
      {
      target = m_Model->PredictBatch(listSample);
      }

    ogr::DataSource::Pointer output;
276 277
    ogr::DataSource::Pointer buffer = ogr::DataSource::New();
    bool updateMode = false;
278 279
    if (IsParameterEnabled("out") && HasValue("out"))
      {
280
      // Create new OGRDataSource
281
      output = ogr::DataSource::New(GetParameterString("out"), ogr::DataSource::Modes::Overwrite);
282 283 284 285 286 287 288 289 290 291 292
      otb::ogr::Layer newLayer = output->CreateLayer(
        GetParameterString("out"),
        const_cast<OGRSpatialReference*>(layer.GetSpatialRef()),
        layer.GetGeomType());
      // Copy existing fields
      OGRFeatureDefn &inLayerDefn = layer.GetLayerDefn();
      for (int k=0 ; k<inLayerDefn.GetFieldCount() ; k++)
        {
        OGRFieldDefn fieldDefn(inLayerDefn.GetFieldDefn(k));
        newLayer.CreateField(fieldDefn);
        }
293 294 295 296
      }
    else
      {
      // Update mode
297
      updateMode = true;
298
      otbAppLogINFO("Update input vector data.");
299 300 301 302 303 304
      // fill temporary buffer for the transfer
      otb::ogr::Layer inputLayer = layer;
      layer = buffer->CopyLayer(inputLayer, std::string("Buffer"));
      // close input data source
      source->Clear();
      // Re-open input data source in update mode
305 306 307
      output = otb::ogr::DataSource::New(shapefile, otb::ogr::DataSource::Modes::Update_LayerUpdate);
      }

308 309
    otb::ogr::Layer outLayer = output->GetLayer(0);

310 311 312 313 314 315
    OGRErr errStart = outLayer.ogr().StartTransaction();
    if (errStart != OGRERR_NONE)
      {
      itkExceptionMacro(<< "Unable to start transaction for OGR layer " << outLayer.ogr().GetName() << ".");
      }

316 317 318 319
    // Add the field of prediction in the output layer if field not exist
    OGRFeatureDefn &layerDefn = layer.GetLayerDefn();
    int idx = layerDefn.GetFieldIndex(GetParameterString("cfield").c_str());
    if (idx >= 0)
320
      {
321 322
      if (layerDefn.GetFieldDefn(idx)->GetType() != OFTInteger)
        itkExceptionMacro("Field name "<< GetParameterString("cfield") << " already exists with a different type!");
323 324
      }
    else
325 326 327
      {
      OGRFieldDefn predictedField(GetParameterString("cfield").c_str(), OFTInteger);
      ogr::FieldDefn predictedFieldDef(predictedField);
328
      outLayer.CreateField(predictedFieldDef);
329
      }
330 331

    // Add confidence field in the output layer
332
    std::string confFieldName("confidence");
333 334
    if (computeConfidenceMap)
      {
335 336 337 338 339 340
      idx = layerDefn.GetFieldIndex(confFieldName.c_str());
      if (idx >= 0)
        {
        if (layerDefn.GetFieldDefn(idx)->GetType() != OFTReal)
          itkExceptionMacro("Field name "<< confFieldName << " already exists with a different type!");
        }
341
      else
342 343 344 345 346
        {
        OGRFieldDefn confidenceField(confFieldName.c_str(), OFTReal);
        confidenceField.SetWidth(confidenceField.GetWidth());
        confidenceField.SetPrecision(confidenceField.GetPrecision());
        ogr::FieldDefn confFieldDefn(confidenceField);
347
        outLayer.CreateField(confFieldDefn);
348
        }
349 350
      }

351
    // Fill output layer
352 353
    unsigned int count=0;
    std::string classfieldname = GetParameterString("cfield");
354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372
    it = layer.cbegin();
    itEnd = layer.cend();
    for( ; it!=itEnd ; ++it, ++count)
      {
      ogr::Feature dstFeature(outLayer.GetLayerDefn());
      dstFeature.SetFrom( *it , TRUE);
      dstFeature.SetFID(it->GetFID());
      dstFeature[classfieldname].SetValue<int>(target->GetMeasurementVector(count)[0]);
      if (computeConfidenceMap)
        dstFeature[confFieldName].SetValue<double>(quality->GetMeasurementVector(count)[0]);
      if (updateMode)
        {
        outLayer.SetFeature(dstFeature);
        }
      else
        {
        outLayer.CreateFeature(dstFeature);
        }
      }
373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396

    if(outLayer.ogr().TestCapability("Transactions"))
      {
      const OGRErr errCommitX = outLayer.ogr().CommitTransaction();
      if (errCommitX != OGRERR_NONE)
        {
        itkExceptionMacro(<< "Unable to commit transaction for OGR layer " << outLayer.ogr().GetName() << ".");
        }
      }

    output->SyncToDisk();

    clock_t toc = clock();
    otbAppLogINFO( "Elapsed: "<< ((double)(toc - tic) / CLOCKS_PER_SEC)<<" seconds.");

  }

  ModelPointerType m_Model;
};

}
}

OTB_APPLICATION_EXPORT(otb::Wrapper::VectorClassifier)