GEE学习笔记八十六：分类中的特征重要性分析

使用GEE做特征重要性分析。

之前在GEE中做随机森林分类时候，很多人都在问如何做特征重要性分析？但是在GEE之前并没有相关API可以做特征重要性分析，最新的API更新后GEE也可以做特征重要性分析了。

1、目前常用的包含特征重要信息分析的分类方法包括：

（1）决策树

ee.Classifier.smileCart(maxNodes, minLeafPopulation)

（2）随机森林

ee.Classifier.smileRandomForest(numberOfTrees, variablesPerSplit, minLeafPopulation, bagFraction, maxNodes, seed)

这两个方法和之前的分类决策树和随机森林分类方法参数非常类似，只不过调用这两个定义方法在使用explain()方法就可以得到每个分类特征重要性的信息。

2、得到特征重要性的API方法：

也就是之前在分类器中的方法explain，这个方法是最开始就有的，只不过是最近GEE官方加入了特征重要性的分析返回信息，下面通过一个具体例子说明一下如何使用以及具体输出。

具体代码：

var roi =
/* color: #d63000 */
/* shown: false */
/* displayProperties: [
{
"type": "rectangle"
}
] */
ee.Geometry.Polygon(
[[[114.23790821489501, 36.43657462800738],
[114.23790821489501, 36.29834769127675],
[114.49265369829345, 36.29834769127675],
[114.49265369829345, 36.43657462800738]]], null, false),
crop =
/* color: #98ff00 */
/* shown: false */
ee.FeatureCollection(
[ee.Feature(
ee.Geometry.Point([114.31343922075439, 36.356156419842804]),
{
"type": 0,
"system:index": "0"
}),
ee.Feature(
ee.Geometry.Point([114.31056389269042, 36.35499859187555]),
{
"type": 0,
"system:index": "1"
}),
ee.Feature(
ee.Geometry.Point([114.31236633714843, 36.3529248270982]),
{
"type": 0,
"system:index": "2"
}),
ee.Feature(
ee.Geometry.Point([114.32159313615966, 36.35461840580203]),
{
"type": 0,
"system:index": "3"
}),
ee.Feature(
ee.Geometry.Point([114.32348141130615, 36.35638107103549]),
{
"type": 0,
"system:index": "4"
}),
ee.Feature(
ee.Geometry.Point([114.32936081346679, 36.35589720612248]),
{
"type": 0,
"system:index": "5"
}),
ee.Feature(
ee.Geometry.Point([114.33515438493896, 36.35565527253804]),
{
"type": 0,
"system:index": "6"
}),
ee.Feature(
ee.Geometry.Point([114.33545479234863, 36.35855842592195]),
{
"type": 0,
"system:index": "7"
}),
ee.Feature(
ee.Geometry.Point([114.29468521532226, 36.35679580999553]),
{
"type": 0,
"system:index": "8"
}),
ee.Feature(
ee.Geometry.Point([114.29979214128662, 36.34922647614173]),
{
"type": 0,
"system:index": "9"
}),
ee.Feature(
ee.Geometry.Point([114.28871998247314, 36.3387180531404]),
{
"type": 0,
"system:index": "10"
}),
ee.Feature(
ee.Geometry.Point([114.28279766496826, 36.33833778756726]),
{
"type": 0,
"system:index": "11"
}),
ee.Feature(
ee.Geometry.Point([114.39553627428222, 36.3299369018726]),
{
"type": 0,
"system:index": "12"
}),
ee.Feature(
ee.Geometry.Point([114.39725288805175, 36.32997147527504]),
{
"type": 0,
"system:index": "13"
}),
ee.Feature(
ee.Geometry.Point([114.39712414201904, 36.32796619257241]),
{
"type": 0,
"system:index": "14"
}),
ee.Feature(
ee.Geometry.Point([114.39918407854248, 36.3291762831369]),
{
"type": 0,
"system:index": "15"
}),
ee.Feature(
ee.Geometry.Point([114.39845451769042, 36.3274475765775]),
{
"type": 0,
"system:index": "16"
}),
ee.Feature(
ee.Geometry.Point([114.40467724260498, 36.32900341420687]),
{
"type": 0,
"system:index": "17"
}),
ee.Feature(
ee.Geometry.Point([114.40480598863769, 36.32734385296429]),
{
"type": 0,
"system:index": "18"
}),
ee.Feature(
ee.Geometry.Point([114.38210377153564, 36.33612529650057]),
{
"type": 0,
"system:index": "19"
}),
ee.Feature(
ee.Geometry.Point([114.38034424242187, 36.336194437797644]),
{
"type": 0,
"system:index": "20"
}),
ee.Feature(
ee.Geometry.Point([114.38098797258544, 36.34041194086619]),
{
"type": 0,
"system:index": "21"
}),
ee.Feature(
ee.Geometry.Point([114.38785442766357, 36.337473540724005]),
{
"type": 0,
"system:index": "22"
}),
ee.Feature(
ee.Geometry.Point([114.38532242235351, 36.34227863160822]),
{
"type": 0,
"system:index": "23"
})]),
nocrop =
/* color: #0b4a8b */
/* shown: false */
ee.FeatureCollection(
[ee.Feature(
ee.Geometry.Point([114.36832794603515, 36.344663782445146]),
{
"type": 1,
"system:index": "0"
}),
ee.Feature(
ee.Geometry.Point([114.37463650163818, 36.345078583829164]),
{
"type": 1,
"system:index": "1"
}),
ee.Feature(
ee.Geometry.Point([114.37742599901367, 36.34576991455905]),
{
"type": 1,
"system:index": "2"
}),
ee.Feature(
ee.Geometry.Point([114.378026813833, 36.34860430638123]),
{
"type": 1,
"system:index": "3"
}),
ee.Feature(
ee.Geometry.Point([114.37386402544189, 36.35009059256129]),
{
"type": 1,
"system:index": "4"
}),
ee.Feature(
ee.Geometry.Point([114.36987289842773, 36.352198996374725]),
{
"type": 1,
"system:index": "5"
}),
ee.Feature(
ee.Geometry.Point([114.37695393022705, 36.351887923993054]),
{
"type": 1,
"system:index": "6"
}),
ee.Feature(
ee.Geometry.Point([114.3797005122583, 36.35392715363849]),
{
"type": 1,
"system:index": "7"
}),
ee.Feature(
ee.Geometry.Point([114.37352070268798, 36.35627738595022]),
{
"type": 1,
"system:index": "8"
}),
ee.Feature(
ee.Geometry.Point([114.31785950121093, 36.34438724696162]),
{
"type": 1,
"system:index": "9"
}),
ee.Feature(
ee.Geometry.Point([114.31657204088378, 36.34345393245458]),
{
"type": 1,
"system:index": "10"
}),
ee.Feature(
ee.Geometry.Point([114.34390911516357, 36.33512274079625]),
{
"type": 1,
"system:index": "11"
}),
ee.Feature(
ee.Geometry.Point([114.35094723161865, 36.33723154988948]),
{
"type": 1,
"system:index": "12"
}),
ee.Feature(
ee.Geometry.Point([114.35309299883056, 36.337058698833104]),
{
"type": 1,
"system:index": "13"
}),
ee.Feature(
ee.Geometry.Point([114.35296425279785, 36.33501902740074]),
{
"type": 1,
"system:index": "14"
}),
ee.Feature(
ee.Geometry.Point([114.41952058729305, 36.35091929703152]),
{
"type": 1,
"system:index": "15"
}),
ee.Feature(
ee.Geometry.Point([114.41966006216182, 36.3516321782695]),
{
"type": 1,
"system:index": "16"
}),
ee.Feature(
ee.Geometry.Point([114.41982099470272, 36.35252219053154]),
{
"type": 1,
"system:index": "17"
}),
ee.Feature(
ee.Geometry.Point([114.42009458002224, 36.35383558809808]),
{
"type": 1,
"system:index": "18"
}),
ee.Feature(
ee.Geometry.Point([114.42027697023525, 36.35499343336394]),
{
"type": 1,
"system:index": "19"
}),
ee.Feature(
ee.Geometry.Point([114.42028233465328, 36.35609509850472]),
{
"type": 1,
"system:index": "20"
}),
ee.Feature(
ee.Geometry.Point([114.42228326257839, 36.35603461533285]),
{
"type": 1,
"system:index": "21"
}),
ee.Feature(
ee.Geometry.Point([114.42357072290554, 36.35606053669799]),
{
"type": 1,
"system:index": "22"
})]),
water =
/* color: #ffc82d */
/* shown: false */
ee.FeatureCollection(
[ee.Feature(
ee.Geometry.Point([114.3321717685144, 36.3306024371752]),
{
"type": 2,
"system:index": "0"
}),
ee.Feature(
ee.Geometry.Point([114.33210739549804, 36.333592954210076]),
{
"type": 2,
"system:index": "1"
}),
ee.Feature(
ee.Geometry.Point([114.28854832109619, 36.41266166922327]),
{
"type": 2,
"system:index": "2"
}),
ee.Feature(
ee.Geometry.Point([114.29610142168212, 36.40879346635716]),
{
"type": 2,
"system:index": "3"
}),
ee.Feature(
ee.Geometry.Point([114.28580173906494, 36.40547771059803]),
{
"type": 2,
"system:index": "4"
}),
ee.Feature(
ee.Geometry.Point([114.29541477617431, 36.40160915003326]),
{
"type": 2,
"system:index": "5"
}),
ee.Feature(
ee.Geometry.Point([114.3057144587915, 36.40547771059803]),
{
"type": 2,
"system:index": "6"
}),
ee.Feature(
ee.Geometry.Point([114.29953464922119, 36.39829308768715]),
{
"type": 2,
"system:index": "7"
}),
ee.Feature(
ee.Geometry.Point([114.39435610231533, 36.3617846267207]),
{
"type": 2,
"system:index": "8"
}),
ee.Feature(
ee.Geometry.Point([114.39429172929897, 36.361179837099506]),
{
"type": 2,
"system:index": "9"
}),
ee.Feature(
ee.Geometry.Point([114.38845524248256, 36.35940000037548]),
{
"type": 2,
"system:index": "10"
}),
ee.Feature(
ee.Geometry.Point([114.38152441438808, 36.35701530096524]),
{
"type": 2,
"system:index": "11"
}),
ee.Feature(
ee.Geometry.Point([114.38133129533901, 36.35653143999478]),
{
"type": 2,
"system:index": "12"
}),
ee.Feature(
ee.Geometry.Point([114.39965614732876, 36.36371992192421]),
{
"type": 2,
"system:index": "13"
}),
ee.Feature(
ee.Geometry.Point([114.40693029817714, 36.3668473954689]),
{
"type": 2,
"system:index": "14"
}),
ee.Feature(
ee.Geometry.Point([114.41094288286342, 36.36890351124372]),
{
"type": 2,
"system:index": "15"
}),
ee.Feature(
ee.Geometry.Point([114.40238127168789, 36.36840244602017]),
{
"type": 2,
"system:index": "16"
})]);
//监督分类
Map.centerObject(roi, 11);
Map.addLayer(roi, {color: "red"}, "roi", false);
var sampleData = crop.merge(nocrop).merge(water);
//Landsat8 SR数据去云
function rmCloud(image) {
var cloudShadowBitMask = (1 << 3);
var cloudsBitMask = (1 << 5);
var qa = image.select("pixel_qa");
var mask = qa.bitwiseAnd(cloudShadowBitMask).eq(0)
.and(qa.bitwiseAnd(cloudsBitMask).eq(0));
return image.updateMask(mask);
}
//缩放
function scaleImage(image) {
var time_start = image.get("system:time_start");
image = image.multiply(0.0001);
image = image.set("system:time_start", time_start);
return image;
}
//NDVI
function NDVI(image) {
return image.addBands(
image.normalizedDifference(["B5", "B4"])
.rename("NDVI"));
}
//NDWI
function NDWI(image) {
return image.addBands(
image.normalizedDifference(["B3", "B5"])
.rename("NDWI"));
}
//NDBI
function NDBI(image) {
return image.addBands(
image.normalizedDifference(["B6", "B5"])
.rename("NDBI"));
}
var l8Col = ee.ImageCollection("LANDSAT/LC08/C01/T1_SR")
.filterBounds(roi)
.filterDate("2018-1-1", "2019-1-1")
.map(rmCloud)
.map(scaleImage)
.map(NDVI)
.map(NDWI)
.map(NDBI);
var bands = [
"B1", "B2", "B3", "B4", "B5", "B6", "B7",
"NDBI", "NDWI", "NDVI"
];
var l8Image = l8Col.median()
.clip(roi)
.select(bands);
var rgbVisParam = {
min: 0,
max: 0.3,
bands: ["B4", "B3", "B2"]
};
Map.addLayer(l8Image, rgbVisParam, "l8Image");
//生成监督分类训练使用的样本数据
var training = l8Image.sampleRegions({
collection: sampleData,
properties: ["type"],
scale: 30
});
//初始化分类器
// var classifier = ee.Classifier.smileCart().train({
// features: training,
// classProperty: "type",
// inputProperties: bands
// });
var classifier = ee.Classifier.smileRandomForest(10).train({
features: training,
classProperty: "type",
inputProperties: bands
});
print(classifier.explain());
//影像数据调用classify利用训练数据训练得到分类结果
var classified = l8Image.classify(classifier);
//训练结果的混淆矩阵
var trainAccuracy = classifier.confusionMatrix();
Map.addLayer(classified, {min:0, max:2, palette: ["green", "orange", "blue"]}, "classify");

代码分析：

（1）这段代码做的是一个简单的监督分类，通过自己标注的样本将影像分成3大类（作物、非作物和水），样本由于是我做测试随机标注的，所以测试精度和分类准确度不高。

（2）和之前的分类代码相比较，这里只是更换了分类方法，比如随机森林我使用的是smileRandomForest方法，参数和之前的方法的参数类似，这里只是简单设置了树的个数为10。

（3）然后对分类器调用explain方法就可以查看特征重要性信息，输出结果如下面所示。

运行结果：

比如使用ee.Classifier.smileRandomForest()方法做的分类结果，使用explain后的信息如下图：

来源请引用：地理遥感生态网科学数据注册与出版系统。

GEE学习笔记 八十六：分类中的特征重要性分析

GEE学习笔记八十六：分类中的特征重要性分析