PIE-engine 教程 ——随机森林监督分类案例（河北安国市为例）

/**
 * @File    :   安国市耕林园地随机森林监督分类
 * @Time    :   2022/3/1
 * @Author  :   piesat
 * @Version :   1.0
 * @Contact :   400-890-0662
 * @License :   (C)Copyright 航天宏图信息技术股份有限公司
 * @Desc    :   安国市耕林园地随机森林监督分类
 */
var imageCollection0 = pie.ImageCollection("GF2/L1A/Fusion");
var featureCollection0 = pie.FeatureCollection("NGCC/CHINA_COUNTY_BOUNDARY");
var imageCollection0 = pie.ImageCollection("NGCC/GLOBELAND30");
// 耕、林、园地分类
// 1.加载研究区-安国市行政边界
var anguo = pie
  .FeatureCollection("NGCC/CHINA_COUNTY_BOUNDARY")
  .filter(pie.Filter.eq("name", "安国市"))
  .first()
  .geometry();
Map.centerObject(anguo, 10);
Map.addLayer(anguo, { color: "red", fillColor: "00000000", width: 2 }, "anguo");
// 2.加载栅格影像
// 数据说明:image为保存于云端、计算好的sentinel合成的NDVI多时序波段数据.
//images_Anguo_City数据中B1、B2、B3为Sentinel的三个波段，B4-B11计算好的sentinel合成的NDVI多时序波段数据
var bands = ["B4", "B5", "B6", "B7", "B8", "B9", "B10", "B11"];
var image = pie
  .Image("user/pieadmin/public/Raster/images_Anguo_City")
  .select(bands)
  .multiply(10000);
print("img", image);
// 加载显示一个波段2020.7.6_NDVI
var image0 = pie.Image("user/pieadmin/public/Raster/images_Anguo_City");
Map.addLayer(image0.select("B4"), { min: -1, max: 1 }, "2020.7.6_NDVI", false);
// Sentinel2数据显示
var image_vis = pie
  .Image("user/pieadmin/public/Raster/images_Anguo_City")
  .select(["B1", "B2", "B3"]);
var visParam = { min: 0, max: 3000, bands: ["B1", "B2", "B3"] };
Map.addLayer(
  image_vis.select(["B1", "B2", "B3"]),
  visParam,
  "Sentinel2_image",
  false
);
// 3.point为样本点，0为耕地，1为林地，2为园地，3为其他地物.
var point = pie.FeatureCollection(
  "user/pieadmin/public/shape/Training_Points_Anguo_City"
);
Map.addLayer(point, { color: "Red" }, "points");
// print(point)
// 获取训练样本
var training = image.sampleRegions(point, ["landcover"], 10, "", "", true);
// Map.addLayer(training, {color: "FAFAD2"}, "trainingpoints");
var withRandom = pie.FeatureCollection("user/HaorenWang/withRandom");
var trainPartition = withRandom.filter(pie.Filter.lte("random", 0.7));
var testPartition = withRandom.filter(pie.Filter.gt("random", 0.7));
Map.addLayer(trainPartition, { color: "red" }, "training");
Map.addLayer(testPartition, { color: "blue" }, "validation");
print("withRandom", withRandom);
print("training", trainPartition);
print("validation", testPartition);
// 4.构建分类器，利用70%数据进行训练.30%数据进行测试
// 构建随机森林Random Forest
var classifer = pie.Classifier.rTrees().train(
  trainPartition,
  "landcover",
  bands
);
// 5.进行分类
var class_image1 = image.classify(classifer);
// 分类后处理：平滑影像，进行开闭运算剔除噪声影响.
var class_image2 = class_image1.focal_min(2).focal_max(2);
// var kernel= pie.Kernel().mean(3)
// class_image2 = class_image1.convolve(kernel)
// 分类结果显示. 蓝色为耕地，绿色为林地，黄色为园地，紫色为其他.
var visParam = {
  opacity: 1,
  uniqueValue: "0,1,2,3",
  palette: "#4169E1,#7CFC00,#FFD700,#FF00FF",
};
//加载分类结果
Map.addLayer(class_image1, visParam, "class_image1");
//加载分类后处理结果
Map.addLayer(class_image2, visParam, "class_image2");
//6.精度验证
var matrix = classifer.confusionMatrix();
print("matrix", matrix);
// 评估训练样本的精度.
var checkM = classifer.confusionMatrix();
print("训练矩阵-ACC系数：", checkM.acc());
print("训练矩阵-Kappa系数：", checkM.kappa());
// 评估验证样本的精度.
var predictResult = testPartition.classify(classifer, "classification");
var errorM = predictResult.errorMatrix("landcover", "classification");
print("验证矩阵-ACC系数：", errorM.acc());
print("验证矩阵-Kappa系数：", errorM.kappa());
// 7.添加图例
var data = {
  title: "图例",
  colors: ["#4169E1", "#7CFC00", "#FFD700", "#FF00FF"],
  labels: ["耕地", "林地", "园地", "其他"],
  step: 1, // 将每个单位色块均分为30个
  isvertical: true,
};
var style = {
  right: "140px", // 图例框的右侧与屏幕最右端的距离
  bottom: "10px", // 图例框的底侧与屏幕底端的距离
  height: "150px", // 图例框的纵向高度
  width: "90px", // 图例框的横向长度
};
var legend = ui.Legend(data, style);
Map.addUI(legend);
//导出影像
Export.image({
  image: class_image2,
  description: "Classified_Image_by_filter",
  region: anguo,
  scale: 10,
});