classification_algorithms

// Loading external scripts.
var altering_IC_algorithms = require('users/albremoteforest/DontDelete:altering_IC_algorithms');


// Function that automatically adapts the band_model to the available bands for a given multitemporal image.
var automatic_band_model_adjust = function(func_mt_image, func_band_model1) {
  var list_number_of_bands = ee.List.sequence(0, func_band_model1.length().subtract(1));
  
  // Analyse MT - image for classification regarding available bands.
  var band_names = func_mt_image.bandNames();
  var length_band_names = ee.Number(band_names.length());
  var letzter_eintrag = ee.String(band_names.get(length_band_names.subtract(1)));
  var index_ = letzter_eintrag.index("_");
  
  // Function that actually builds the custom band model.
  var construct_band_model_custom = function(last_item, index_of_, func_bands) {
    // Extracts the number of single images used in constructing the multitemporal image.
    var einzelbilder = ee.Number.parse(letzter_eintrag.slice(index_.add(1), index_.add(2)));
    var list_number_of_images = ee.List.sequence(1, einzelbilder);
    // Prepare band_model to match bands of image for classification.
    // Construct band names (band names for classification + _ + number of single pictures in multitemporal image.
    var band_model1 = list_number_of_images.map(function(number) {
      number = ee.Number(number).toInt();
      number = ee.String(number);
      var wrap = func_bands.map(function(band) {
        band = ee.String(band);
        var band_plus_number = band.cat("_").cat(number);
        return band_plus_number;
      });
      return wrap;
    });
  
    // Append all lists together.
    var final_list = list_number_of_images.iterate(
      function(instance, list) {
        instance = ee.Number(instance).subtract(1);
        list = ee.List(list);
        var temp_list = list.add(band_model1.get(instance));
        return temp_list;
      }, 
      ee.List(func_bands)
    );
    return ee.List(final_list).flatten();
  };

  // If-statement, activates function construct_band_model_custom only if a multitemporal image is in use.
  // Otherwise original band_model is assumed. 
  var band_model_custom = ee.Algorithms.If(
    ee.Algorithms.IsEqual(index_, -1),
    band_model_custom = func_band_model1,
    band_model_custom = construct_band_model_custom(letzter_eintrag, index_, func_band_model1)
  );
  
 return band_model_custom;
};


// Function that classifies one image, based on the training data passed to this function and a Random Forest.
var classifyImage = function(func_image, func_training_samples, func_band_model) {
  func_image = ee.Image(func_image);
  func_training_samples = ee.FeatureCollection(func_training_samples);
  var training_properties = func_training_samples.get('classification_feature');
  var func_band_model_custom = automatic_band_model_adjust(func_image, func_band_model);
  
  // Sampling training data.
  var training = func_image.select(func_band_model_custom).sampleRegions({
    collection: func_training_samples, 
    properties: [training_properties], 
    scale: 10
  });
  
  // Train classifier.
  var classifier = ee.Classifier.randomForest(500, 0).train({
    features: training,
    classProperty: training_properties,
    inputProperties: func_band_model_custom
  });
  
  // Classify image.
  var classified_image = func_image.select(func_band_model_custom).classify(classifier);
  
  // Explain you, classifier!
  var clas_explain = classifier.explain();
  
  // Combine classified picture and explain into dictionary for return.
  var dic_img_expl = ee.Dictionary({
    'classified_image': classified_image,
    'explanation': clas_explain
  });
  
  return classified_image;
  //return dic_img_expl;
};


// Function that classifies one image, based on the training data passed to this function and a Random Forest, 
// but returns the probability of the classification.
var classification_probability = function (func_image, func_training_samples, func_band_model) {
  func_image = ee.Image(func_image);
  func_training_samples = ee.FeatureCollection(func_training_samples);
  var training_properties = func_training_samples.get('classification_feature');
  var func_band_model_custom = automatic_band_model_adjust(func_image, func_band_model);
  
  // Sampling training data.
  var training = func_image.select(func_band_model_custom).sampleRegions({
    collection: func_training_samples, 
    properties: [training_properties], 
    scale: 10
  });
  
  // Train classifier.
  var classifier = ee.Classifier.randomForest(500, 0).setOutputMode('PROBABILITY').train({
    features: training,
    classProperty: training_properties,
    inputProperties: func_band_model_custom
  });
  
  // Classify image.
  var classified_image = func_image.select(func_band_model_custom).classify(classifier);
  
  return classified_image;
};


// Function that takes a year, loads predefined images, combines them into one and classifies them in two steps.
// First deciduous and coniferous, then ash, beech and mixed forest. This basic forest classification is used to compare 
// single-date classifiactions with.
var classifyForestMain = function(year, sample_data) {
  year = ee.String(year);
  sample_data = ee.Dictionary(sample_data);
  var sample_data_classification_step_2 = ee.FeatureCollection(sample_data.get('dec_vs_con'));
  var sample_data_classification_step_3 = ee.FeatureCollection(sample_data.get('ash_vs_beech'));
  var bands_step_2 = ee.List(['B3', 'B4', 'B8']);
  var bands_step_3 = ee.List(['B3', 'B4', 'B8']);  
  var func_bands = ee.List(['B2', 'B3', 'B8', 'EVI', 'NDVI']);
  var forest_mask = ee.Image('users/albremoteforest/forest_masks_2016---2019')
  .select(ee.String('FM_').cat(year))
  .focal_min({kernel: ee.Kernel.circle(1)}).focal_max({kernel: ee.Kernel.circle(1)});
  
  var combine_2_images = function(img1, img2) {
    img1 = ee.Image(altering_IC_algorithms.adding_ndvi_evi(img1));
    img2 = ee.Image(altering_IC_algorithms.adding_ndvi_evi(img2));
    var combined_image = img1.addBands(img2);
    return combined_image;
  };
  
  // Loads 3 or 4 pictures per year and combines them to one picture.
  var img_classify_step2 = ee.Algorithms.If(
    ee.Algorithms.IsEqual(year, '2016'),
    img_classify_step2 = combine_2_images(
      ee.Image('COPERNICUS/S2/20160505T103032_20160505T103027_T32UNU'),
      ee.Image('COPERNICUS/S2/20160823T103022_20160823T103332_T32UNU')),
    img_classify_step2 = ee.Algorithms.If(
      ee.Algorithms.IsEqual(year, '2017'),
      img_classify_step2 = combine_2_images(
        ee.Image('COPERNICUS/S2_SR/20170517T102031_20170517T102352_T32UNU'),
        ee.Image('COPERNICUS/S2_SR/20170626T102021_20170626T102321_T32UNU')),
      img_classify_step2 = ee.Algorithms.If(
        ee.Algorithms.IsEqual(year, '2018'),
        img_classify_step2 = combine_2_images(
          ee.Image('COPERNICUS/S2_SR/20180427T102019_20180427T102022_T32UNU'),
          ee.Image('COPERNICUS/S2_SR/20180701T102021_20180701T102404_T32UNU')),
        img_classify_step2 = ee.Algorithms.If(
          ee.Algorithms.IsEqual(year, '2019'),
          img_classify_step2 = combine_2_images(
            ee.Image('COPERNICUS/S2_SR/20190517T102031_20190517T102508_T32UNU'),
            ee.Image('COPERNICUS/S2_SR/20190629T103031_20190629T103537_T32UNU')),
          img_classify_step2 = ee.Image([])
        )
      )
    )  
  );
  img_classify_step2 = ee.Image(img_classify_step2).updateMask(forest_mask).set('year', year);

  // Classify deciduous and coniferous forest without explain.
  var img_classified_dec_con = classifyImage(img_classify_step2, sample_data_classification_step_2, bands_step_2)
    .set('classification_step:', 'deciduous_vs_coniferous', 'year:', year);
  // Calculate classification probability for above classification step (step2).
  var img_classified_dec_con_prob = classification_probability(img_classify_step2, sample_data_classification_step_2, bands_step_2)
    .set('classification_step:', 'deciduous_vs_coniferous_probability', 'year:', year);

  // Classify deciduous and coniferous forest with explain.
  //var img_classified_dec_con = classifyImage(img_classify_step2, sample_data_classification_step_2, bands_step_2);
  
  // Classify ash and beech trees within previously classified deciduous forest areas without explain.
  var img_classify_step3 = img_classify_step2.updateMask(
    img_classified_dec_con.remap([0, 1], [1, 0]));
  var img_classified_ash_beech = classifyImage(img_classify_step3, sample_data_classification_step_3, bands_step_3)
    .set('classification_step:', 'ash_vs_beech', 'year:', year);
  // Calculate classificaction probability for above classification step (step_3).
  var img_classified_ash_beech_prob = classification_probability(img_classify_step3, sample_data_classification_step_3, bands_step_3)
    .set('classification_step:', 'ash_vs_beech_probability', 'year:', year);
  
  // Classify ash and beech trees within previously classified deciduous forest areas with explain.
  //var img_classified_dec_con_img = ee.Image(img_classified_dec_con.get('classified_image'));
  //var img_classify_step3 = img_classify_step2.updateMask(
  //  img_classified_dec_con_img.remap([0, 1], [1 ,0]));
  //var img_classified_ash_beech = classifyImage(img_classify_step3, sample_data_classification_step_3, bands_step_3);
  
  // Combine classification results into dictionary for returning, without explain.
  var classification_step_2_and_3 = ee.Dictionary({
    'classification_step_2_classification': ee.Image(img_classified_dec_con),
    'classification_step_2_probability': ee.Image(img_classified_dec_con_prob),
    'classification_step_3_classification': ee.Image(img_classified_ash_beech),
    'classification_step_3_probability': ee.Image(img_classified_ash_beech_prob)
  });
  
  // Combine classification results into dictionary for returning, with explain.
  //var classification_step_2_step_3 = ee.Dictionary({
  //  'classification_step_2': ee.Dictionary(img_classified_dec_con),
  //  'classification_step_3': ee.Dictionary(img_classified_ash_beech)
  //});
  
  
  //return ee.Image(img_classified_ash_beech.get('classified_image'));  
  return classification_step_2_and_3;
};

exports.automatic_band_model_adjust = automatic_band_model_adjust;
exports.classifyImage = classifyImage;
exports.classification_probability = classification_probability;
exports.classifyForestMain = classifyForestMain;