google earth engine随缘学习（七）基于对象分类

这里以前用过的代码，今天贴出来，当初是想提取水体后分出河流和鱼塘，但是分类效果并不是特别理想，以后想想用别的方法，用的哨兵数据，包含一个去云算法。

第一步：去云并提取水体

function sentinel2toa(img) {
  var toa = img.select(['B1','B2','B3','B4','B5','B6','B7','B8','B8A','B9','B10', 'B11','B12'],  
                       ['aerosol', 'blue', 'green', 'red', 're1','re2','re3', 'nir','nir2', 'h2o', 'cirrus','swir1', 'swir2'])
                       .divide(10000)
                       .addBands(img.select(['QA60']))
                       .set('solar_azimuth',img.get('MEAN_SOLAR_AZIMUTH_ANGLE'))
                       .set('solar_zenith',img.get('MEAN_SOLAR_ZENITH_ANGLE'))
    return toa;
}
//-------------------------------cloud mask----------------------------------------------//
function ESAcloud(toa) {
  var qa = toa.select('QA60');
  var cloudBitMask = Math.pow(2, 10);
  var cirrusBitMask = Math.pow(2, 11);
  var clear = qa.bitwiseAnd(cloudBitMask).eq(0).and(
             qa.bitwiseAnd(cirrusBitMask).eq(0));
  var cloud = clear.eq(0);
  return cloud;
}
function shadowMask(toa,cloud){
  var azimuth =ee.Number(toa.get('solar_azimuth')).multiply(Math.PI).divide(180.0).add(ee.Number(0.5).multiply(Math.PI));
  var zenith  =ee.Number(0.5).multiply(Math.PI ).subtract(ee.Number(toa.get('solar_zenith')).multiply(Math.PI).divide(180.0));
  var nominalScale = cloud.projection().nominalScale();
  var cloudHeights = ee.List.sequence(200,10000,500);
  var shadows = cloudHeights.map(function(cloudHeight){
    cloudHeight = ee.Number(cloudHeight);
    var shadowVector = zenith.tan().multiply(cloudHeight);
    var x = azimuth.cos().multiply(shadowVector).divide(nominalScale).round();
    var y = azimuth.sin().multiply(shadowVector).divide(nominalScale).round();
    return cloud.changeProj(cloud.projection(), cloud.projection().translate(x, y));
  });
  var potentialShadow = ee.ImageCollection.fromImages(shadows).max();
  potentialShadow = potentialShadow.and(cloud.not());
  var darkPixels = toa.normalizedDifference(['green', 'swir2']).gt(0.25).rename(['dark_pixels']);
  var shadow = potentialShadow.and(darkPixels).rename('shadows');
  return shadow;
}

function cloud_and_shadow_mask(img) {
  var toa = sentinel2toa(img);
  var cloud = ESAcloud(toa);
  var shadow = shadowMask(toa,cloud);
  var mask = cloud.or(shadow).eq(0);
  return toa.updateMask(mask);
}
//-------------------------------Main----------------------------------------------------------------------------------------//
var s2 = ee.ImageCollection('COPERNICUS/S2');
var image = s2.filterDate('2016-01-01', '2016-12-31')
                  // Pre-filter to get less cloudy granules.
                  .filterBounds(table)
                  .map(cloud_and_shadow_mask)
                  .mean()
                  .clip(table);
var MNDWI =sen.expression('float(2*green-nir-swir)/(2*green+nir+swir)',{'green':sen.select('green'),
'nir':sen.select('nir'),'swir':sen.select('swir1')});
var water=MNDWI.gt(0.1);

第二步：创建几何特征并用随机森林分类

var img=water;
var patchid = img.connectedComponents(ee.Kernel.square(1), 256);
Map.addLayer(patchid.select("labels").randomVisualizer(), {}, 'patches',false);   //显示各个斑块
var minMax = img.reduceNeighborhood({
            reducer: ee.Reducer.minMax(), 
             kernel: ee.Kernel.square(1)
}); 
var area = ee.Image.pixelArea().addBands(img)   //面积
    .reduceConnectedComponents(ee.Reducer.sum(), 'constant', 256);
var perimeterPixels = minMax.select(0).neq(minMax.select(1));   
var perimeter = perimeterPixels.addBands(img) //周长
    .reduceConnectedComponents(ee.Reducer.sum(),'constant',256);
var sizes = ee.Image.pixelLonLat().addBands(img)
    .reduceConnectedComponents(ee.Reducer.minMax(),'constant', 256);
var width = sizes.select('longitude_max')  //宽度
    .subtract(sizes.select('longitude_min')).rename('width');
var height = sizes.select('latitude_max') //长度
    .subtract(sizes.select('latitude_min')).rename('height');
var LSI = perimeter.multiply(0.25).divide((area.sqrt()));  //形状指数
var stdDev = image2.addBands(img)  //方差
    .reduceConnectedComponents(ee.Reducer.stdDev(),'constant', 256);
var train = [fishh,river];  //选择训练样本
var TrainingSamples = ee.FeatureCollection(train);
var objectPropertiesImage = ee.Image.cat([
  image,
  area,
  perimeter,
  width,
  height,
  LSI
]).float();
var training = objectPropertiesImage.sampleRegions({
    collection: TrainingSamples,
  properties: ['class'],
  scale: 10
});


var classifier = ee.Classifier.randomForest(100,0,1,0.5,true).train(training, 'class');
var classified = objectPropertiesImage.classify(classifier);  //分类
Map.addLayer(classified,{},'classified');

如果基于对象分类配合图像分割算法的话可能会比单纯的阈值分割更好一点，但是那个图像分割的结果不知道怎么用惹~