遥感影像的面向对象方法

//-------------------------------去云  ----------------------------------------------//
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;
}
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 table = ee.FeatureCollection("users/xzhangsx/sample200");
var bands =['B1','B2', 'B3', 'B4', 'B5', 'B6','B7','B8','B8A','B9'];
var composite = ee.ImageCollection('COPERNICUS/S2')
    .filterDate('2020-06-01', '2020-06-15')
    .filterBounds(table)
    .mean();
var image = ee.Image(composite).divide(255).select(bands);
print(image);
//设置种子
var seeds = ee.Algorithms.Image.Segmentation.seedGrid(36);
//利用 SNIC 进行分割，得到对象。参数的设置根据自己的需求来
var snic = ee.Algorithms.Image.Segmentation.SNIC({
  image: image, 
  size: 100,
  compactness: 5,
  connectivity: 8,
  neighborhoodSize:256,
  seeds: seeds
}).select(['B1_mean','B2_mean','B3_mean','B4_mean','B5_mean','B6_mean','B7_mean','B8_mean','B8A_mean','B9_mean', 'clusters'], ['B1','B2','B3','B4','B5','B6','B7','B8','B8A','B9', 'clusters']);
var clusters = snic.select('clusters');//
print("snic",snic);
//计算每个对象的特征
var stdDev = image.addBands(clusters).reduceConnectedComponents(ee.Reducer.stdDev(), 'clusters', 256);//为啥和上面的均值不一样？
var area = ee.Image.pixelArea().addBands(clusters).reduceConnectedComponents(ee.Reducer.sum(), 'clusters', 256);
var minMax = clusters.reduceNeighborhood(ee.Reducer.minMax(), ee.Kernel.square(1));
var perimeterPixels = minMax.select(0).neq(minMax.select(1)).rename('perimeter');
var perimeter = perimeterPixels.addBands(clusters).reduceConnectedComponents(ee.Reducer.sum(), 'clusters', 256);
var sizes = ee.Image.pixelLonLat().addBands(clusters).reduceConnectedComponents(ee.Reducer.minMax(), 'clusters', 256);
var width = sizes.select('longitude_max').subtract(sizes.select('longitude_min')).rename('width');
var height = sizes.select('latitude_max').subtract(sizes.select('latitude_min')).rename('height');
//分类器参数设置，选择分类依据：包括了area，width等
var objectPropertiesImage = ee.Image.cat([
  snic.select(bands),
  //stdDev,
 // area,
  //perimeter,
  //width,
  //height
]).float();
print(objectPropertiesImage);
//var img2 = ee.Image(objectPropertiesImage).divide(255);
//选择训练样本
var training = objectPropertiesImage.sampleRegions({
  collection: table,
  properties: ['value'],
  scale: 30,
  tileScale:8
});
print("training",training);
 
/*var training = objectPropertiesImage.addBands(table.select('value'))
    .updateMask(seeds)
    .sample(table, 5);
print("training2",training2);*/
//分类
var classifier = ee.Classifier.smileCart().train(training, 'value');
print(classifier);
var result=objectPropertiesImage.classify(classifier,'Classified objects');
//Export.image.toCloudStorage(result,"202012172236");
//var resultshp = result.updateMask(seeds).sample(table, 5);
//Export.table.toCloudStorage(resultshp,"202012172120");
print(result);
Map.centerObject(table);
Map.addLayer(result.randomVisualizer(), {}, 'Classified objects2');
// 计算混淆矩阵
var test=training.classify(classifier,'classification');
var confusionMatrix = test.errorMatrix('value', 'classification');
print('Confusion Matrix', confusionMatrix);

// Import a high-resolution Landsat 8 image
var image = ee.Image('LANDSAT/LC08/C01/T1_SR/LC08_044034_20140318');

// Define the SNIC parameters
var size = 32; // the seed distance in pixels
var compactness = 5; // the segmentation compactness factor
var connectivity = 8; // the connectivity mode (4 or 8)
var neighborhoodSize = 256; // the neighborhood size in pixels

// Apply the SNIC algorithm to the image
var snic = ee.Algorithms.Image.Segmentation.SNIC({
  image: image, 
  size: size,
  compactness: compactness,
  connectivity: connectivity,
  neighborhoodSize: neighborhoodSize
});

// Display the original image and the segmentation result
Map.centerObject(image, 10);
Map.addLayer(image, {bands: ['B4', 'B3', 'B2'], min: 0, max: 3000}, 'Original Image');
Map.addLayer(snic.randomVisualizer(), {}, 'Segmentation Result');

// 1.1 Unsupervised k-Means classification
 
// This function does unsupervised clustering classification 
// input = any image. All bands will be used for clustering.
// numberOfUnsupervisedClusters = tunable parameter for how 
//        many clusters to create.
var afn_Kmeans = function(input, numberOfUnsupervisedClusters,
    defaultStudyArea, nativeScaleOfImage) {
 
    // Make a new sample set on the input. Here the sample set is 
    // randomly selected spatially. 
    var training = input.sample({
        region: defaultStudyArea,
        scale: nativeScaleOfImage,
        numPixels: 1000
    });
 
    var cluster = ee.Clusterer.wekaKMeans(
            numberOfUnsupervisedClusters)
        .train(training);
 
    // Now apply that clusterer to the raw image that was also passed in. 
    var toexport = input.cluster(cluster);
 
    // The first item is the unsupervised classification. Name the band.
    var clusterUnsup = toexport.select(0).rename(
        'unsupervisedClass');
    return (clusterUnsup);
};

// 1.2 Simple normalization by maxes function.
var afn_normalize_by_maxes = function(img, bandMaxes) {
    return img.divide(bandMaxes);
};
 
// 1.4 Simple add mean to Band Name function
var afn_addMeanToBandName = (function(i) {
    return i + '_mean';
});

 // 2. Parameters to function calls
 
 
// 2.1. Unsupervised KMeans Classification Parameters
var numberOfUnsupervisedClusters = 4;

 // 2.2. Visualization and Saving parameters
// For different images, you might want to change the min and max 
// values to stretch. Useful for images 2 and 3, the normalized images.
var centerObjectYN = true;

//
// 3. Statements
//
 
// 3.1  Selecting Image to Classify 
var whichImage = 1; // will be used to select among images
if (whichImage == 1) {
    // Image 1. 
    // Puget Sound, WA: Forest Harvest
    // (April 21, 2016)
    // Harvested Parcels
    // Clear Parcel Boundaries
    // Sentinel 2, 10m
    var whichCollection = 'COPERNICUS/S2';
    var ImageToUseID = '20160421T191704_20160421T212107_T10TDT';
    var originalImage = ee.Image(whichCollection + '/' +
    ImageToUseID);
    print(ImageToUseID, originalImage);
    var nativeScaleOfImage = 10;
    var threeBandsToDraw = ['B4', 'B3', 'B2'];
    var bandsToUse = ['B4', 'B3', 'B2'];
    var bandMaxes = [1e4, 1e4, 1e4];
    var drawMin = 0;
    var drawMax = 0.3;
    var defaultStudyArea = ee.Geometry.Polygon(
        [
            [
                [-123.13105468749993, 47.612974066532004],
                [-123.13105468749993, 47.56214700543596],
                [-123.00179367065422, 47.56214700543596],
                [-123.00179367065422, 47.612974066532004]
            ]
        ]);
    var zoomArea = ee.Geometry.Polygon(
        [
            [
                [-123.13105468749993, 47.612974066532004],
                [-123.13105468749993, 47.56214700543596],
                [-123.00179367065422, 47.56214700543596],
                [-123.00179367065422, 47.612974066532004]
            ]
        ], null, false);
}
Map.addLayer(originalImage.select(threeBandsToDraw), {
    min: 0,
    max: 2000
}, '3.1 ' + ImageToUseID, true, 1);

// 4. Image Preprocessing 
 
var clippedImageSelectedBands = originalImage.clip(defaultStudyArea)
    .select(bandsToUse);
var ImageToUse = afn_normalize_by_maxes(clippedImageSelectedBands,
    bandMaxes);
 
Map.addLayer(ImageToUse.select(threeBandsToDraw), {
        min: 0.028,
        max: 0.12
    },
    '4.3 Pre-normalized image', true, 0);

// 6. Execute Classifications
 
 
// 6.1 Per Pixel Unsupervised Classification for Comparison
var PerPixelUnsupervised = afn_Kmeans(ImageToUse,
    numberOfUnsupervisedClusters, defaultStudyArea,
    nativeScaleOfImage);
Map.addLayer(PerPixelUnsupervised.select('unsupervisedClass')
    .randomVisualizer(), {}, '6.1 Per-Pixel Unsupervised', true, 0
);
print('6.1b Per-Pixel Unsupervised Results:', PerPixelUnsupervised);

// 7. Zoom if requested
 
if (centerObjectYN === true) {
    Map.centerObject(zoomArea, 14);
}

// 1.3 Seed Creation and SNIC segmentation Function
var afn_SNIC = function(imageOriginal, SuperPixelSize, Compactness,
    Connectivity, NeighborhoodSize, SeedShape) {
    var theSeeds = ee.Algorithms.Image.Segmentation.seedGrid(
        SuperPixelSize, SeedShape);
    var snic2 = ee.Algorithms.Image.Segmentation.SNIC({
        image: imageOriginal,
        size: SuperPixelSize,
        compactness: Compactness,
        connectivity: Connectivity,
        neighborhoodSize: NeighborhoodSize,
        seeds: theSeeds
    });
    var theStack = snic2.addBands(theSeeds);
    return (theStack);
};

// 2.3 Object-growing parameters to change
// Adjustable Superpixel Seed and SNIC segmentation Parameters:
// The superpixel seed location spacing, in pixels.
var SNIC_SuperPixelSize = 16;
// Larger values cause clusters to be more compact (square/hexagonal). 
// Setting this to 0 disables spatial distance weighting.
var SNIC_Compactness = 0;
// Connectivity. Either 4 or 8. 
var SNIC_Connectivity = 4;
// Either 'square' or 'hex'.
var SNIC_SeedShape = 'square';
 
// 2.4 Parameters that can stay unchanged
// Tile neighborhood size (to avoid tile boundary artifacts). Defaults to 2 * size.
var SNIC_NeighborhoodSize = 2 * SNIC_SuperPixelSize;

// 5. SNIC Clustering
 
 
// This function returns a multi-banded image that has had SNIC
// applied to it. It automatically determine the new names 
// of the bands that will be returned from the segmentation.
print('5.1 Execute SNIC');
var SNIC_MultiBandedResults = afn_SNIC(
    ImageToUse,
    SNIC_SuperPixelSize,
    SNIC_Compactness,
    SNIC_Connectivity,
    SNIC_NeighborhoodSize,
    SNIC_SeedShape
);
 
var SNIC_MultiBandedResults = SNIC_MultiBandedResults
    .reproject('EPSG:3857', null, nativeScaleOfImage);
print('5.2 SNIC Multi-Banded Results', SNIC_MultiBandedResults);
 
Map.addLayer(SNIC_MultiBandedResults.select('clusters')
    .randomVisualizer(), {}, '5.3 SNIC Segment Clusters', true, 1);
 
var theSeeds = SNIC_MultiBandedResults.select('seeds');
Map.addLayer(theSeeds, {
    palette: 'red'
}, '5.4 Seed points of clusters', true, 1);
 
var bandMeansToDraw = threeBandsToDraw.map(afn_addMeanToBandName);
print('5.5 band means to draw', bandMeansToDraw);
var clusterMeans = SNIC_MultiBandedResults.select(bandMeansToDraw);
print('5.6 Cluster Means by Band', clusterMeans);
Map.addLayer(clusterMeans, {
    min: drawMin,
    max: drawMax
}, '5.7 Image repainted by segments', true, 0);

// The superpixel seed location spacing, in pixels.
var SNIC_SuperPixelSize = 16;
// Larger values cause clusters to be more compact (square/hexagonal). 
// Setting this to 0 disables spatial distance weighting.
var SNIC_Compactness = 0;
// Connectivity. Either 4 or 8. 
var SNIC_Connectivity = 4;
// Either 'square' or 'hex'.
var SNIC_SeedShape = 'square';

// 6.2 SNIC Unsupervised Classification for Comparison
var bandMeansNames = bandsToUse.map(afn_addMeanToBandName);
print('6.2 band mean names returned by segmentation', bandMeansNames);
var meanSegments = SNIC_MultiBandedResults.select(bandMeansNames);
var SegmentUnsupervised = afn_Kmeans(meanSegments,
    numberOfUnsupervisedClusters, defaultStudyArea,
    nativeScaleOfImage);
Map.addLayer(SegmentUnsupervised.randomVisualizer(), {},
    '6.3 SNIC Clusters Unsupervised', true, 0);
print('6.3b Per-Segment Unsupervised Results:', SegmentUnsupervised);