Export.classifier.toAsset

Cria uma tarefa em lote para exportar um ee.Classifier como um recurso do Earth Engine.

Suporte apenas para ee.Classifier.smileRandomForest, ee.Classifier.smileCart, ee.Classifier.DecisionTree e ee.Classifier.DecisionTreeEnsemble.

Uso	Retorna
Export.classifier.toAsset(classifier, description, assetId, priority)

Argumento	Tipo	Detalhes
classifier	ComputedObject	O classificador a ser exportado.
description	String, opcional	Um nome legível da tarefa. O padrão é "myExportClassifierTask".
assetId	String, opcional	O ID do recurso de destino.
priority	Número, opcional	A prioridade da tarefa no projeto. As tarefas de maior prioridade são programadas mais cedo. Precisa ser um número inteiro entre 0 e 9999. O padrão é 100.

Exemplos

Editor de código (JavaScript)

// First gather the training data for a random forest classifier.
// Let's use MCD12Q1 yearly landcover for the labels.
var landcover = ee.ImageCollection('MODIS/061/MCD12Q1')
    .filterDate('2022-01-01', '2022-12-31')
    .first()
    .select('LC_Type1');
// A region of interest for training our classifier.
var region = ee.Geometry.BBox(17.33, 36.07, 26.13, 43.28);

// Training features will be based on a Landsat 8 composite.
var l8 = ee.ImageCollection('LANDSAT/LC08/C02/T1')
  .filterBounds(region)
    .filterDate('2022-01-01', '2023-01-01');

// Draw the Landsat composite, visualizing true color bands.
var landsatComposite = ee.Algorithms.Landsat.simpleComposite({
  collection: l8,
  asFloat: true
});
Map.addLayer(landsatComposite, {
  min: 0,
  max: 0.3,
  bands: ['B3', 'B2', 'B1']
}, 'Landsat composite');

// Make a training dataset by sampling the stacked images.
var training = landcover.addBands(landsatComposite).sample({
  region: region,
  scale: 30,
  // With export to Classifier we can bump this higher to say 10,000.
  numPixels: 1000
});

var classifier = ee.Classifier.smileRandomForest({
  // We can also increase the number of trees higher to ~100 if needed.
  numberOfTrees: 3
}).train({features: training, classProperty: 'LC_Type1'});

// Create an export classifier task to run.
var assetId = 'projects/<project-name>/assets/<asset-name>';  // <> modify these
Export.classifier.toAsset({
  classifier: classifier,
  description: 'classifier_export',
  assetId: assetId
});

// Load the classifier after the export finishes and visualize.
var savedClassifier = ee.Classifier.load(assetId)
var landcoverPalette = '05450a,086a10,54a708,78d203,009900,c6b044,dcd159,' +
  'dade48,fbff13,b6ff05,27ff87,c24f44,a5a5a5,ff6d4c,69fff8,f9ffa4,1c0dff';
var landcoverVisualization = {
  palette: landcoverPalette,
  min: 0,
  max: 16,
  format: 'png'
};
Map.addLayer(
    landsatComposite.classify(savedClassifier),
    landcoverVisualization,
    'Upsampled landcover, saved');

Configuração do Python

Consulte a página Ambiente Python para informações sobre a API Python e o uso de geemap para desenvolvimento interativo.

import ee
import geemap.core as geemap

Colab (Python)

# First gather the training data for a random forest classifier.
# Let's use MCD12Q1 yearly landcover for the labels.
landcover = (ee.ImageCollection('MODIS/061/MCD12Q1')
             .filterDate('2022-01-01', '2022-12-31')
             .first()
             .select('LC_Type1'))

# A region of interest for training our classifier.
region = ee.Geometry.BBox(17.33, 36.07, 26.13, 43.28)

# Training features will be based on a Landsat 8 composite.
l8 = (ee.ImageCollection('LANDSAT/LC08/C02/T1')
      .filterBounds(region)
      .filterDate('2022-01-01', '2023-01-01'))

# Draw the Landsat composite, visualizing true color bands.
landsatComposite = ee.Algorithms.Landsat.simpleComposite(
    collection=l8, asFloat=True)

Map = geemap.Map()
Map  # Render the map in the notebook.
Map.addLayer(landsatComposite, {
    'min': 0,
    'max': 0.3,
    'bands': ['B3', 'B2', 'B1']
}, 'Landsat composite')

# Make a training dataset by sampling the stacked images.
training = landcover.addBands(landsatComposite).sample(
    region=region,
    scale=30,
    # With export to Classifier we can bump this higher to say 10,000.
    numPixels=1000
)

# We can also increase the number of trees higher to ~100 if needed.
classifier = ee.Classifier.smileRandomForest(
    numberOfTrees=3).train(features=training, classProperty='LC_Type1')

# Create an export classifier task to run.
asset_id = 'projects/<project-name>/assets/<asset-name>'  # <> modify these
ee.batch.Export.classifier.toAsset(
    classifier=classifier,
    description='classifier_export',
    assetId=asset_id
)

# Load the classifier after the export finishes and visualize.
savedClassifier = ee.Classifier.load(asset_id)
landcover_palette = [
    '05450a', '086a10', '54a708', '78d203', '009900',
    'c6b044', 'dcd159', 'dade48', 'fbff13', 'b6ff05',
    '27ff87', 'c24f44', 'a5a5a5', 'ff6d4c', '69fff8',
    'f9ffa4', '1c0dff']
landcoverVisualization = {
    'palette': landcover_palette,
    'min': 0,
    'max': 16,
    'format': 'png'
}
Map.addLayer(
    landsatComposite.classify(savedClassifier),
    landcoverVisualization,
    'Upsampled landcover, saved')