《精度评估步骤》由会员分享，可在线阅读，更多相关《精度评估步骤（6页珍藏版）》请在金锄头文库上搜索。
1、Uuijiui Pnmt Tcaiiire Class pr sridwiplOutpui Loca.liE. d&iLggu.arsCarcel或町 Hide HelpOKEnyranments. ICcoiEtT-aiiLiii.p;pti di泌;毎 Layers日ra*屯dgl著 C.o pyX kemQ/eType in the keyword to find:nT| Open Attribute TableJoins and RelatesJoin.金 Zoom To Layer 蠢 Zoom To Make VisibleVisible Scale RangeRe.move Jo
2、inCs)Relate.Remote Re I ate (s)ArcMap 中进行精度评估的步骤，以东莞市为例1. 根据东莞市的边界，在区域内生成离散点，每个类设置20 个样点，供设置200 个样本点八、护 Creale Random PointsNlumber Poiimts value orfleld optional)Tha number of paims to bo randomly placed. When a single number is specified, each Feature in the future class will hai/e Ihs samE numb已o
3、F 時ndornly placed points. Whan a fie Id is spe匚iifedi the number of randomlv placed poinls is equal lo thelue of the field 5pe 匚iiiEd fin that featu巳Tool Hei2. 提取样本点所在图斑的地类属性Convert Lbek tQ Annotation. 争 Convert Featuw to Graphics.Convert Symbology to R.epr畐百ent曰tiori.”會 Properties.De-Sac Masks cart
4、ographyJoin DataS3Join lets you append additional data to this layers attribute table so you can, for example fymbdize the layers features using this dmt乩What do you want to join to this layer?Jain data From another layer ban pahial location1. Choose the layer to join to this layerj or lad spatial d
5、ata from disk;2. V&u are joining! Polygons to PointsSelect a join Feature class above. Vou will be given diFFerent options tiaeed on geometry types of the source feature class and the join feature class.Emch paint will be given all the attributes af the polygon that:毡 it falls inside.If 0 point fall
6、s inside more than one polygon (For examplej be匚ause the layer being joined contains overlapping polygons) the attributes of the first polygon found 观ill be joined.is closest to it.A distance Field is added showing how close the polygon is (in the units of the target layer). A polygon that the point
7、 Falls inside is treated as being closest ta the point (i.e. a distance oF 0).3. The result of the join will be saved into new layer.Specify output shapefile or feature class for this new layer:E :dongguanrandompoint.shpAbout Joining DataOKCancel3. 在生成的具有地类属性的样本点中，新建一个class_num字段，字段的值通过select by att
8、ribute 分别赋值，耕地=1，园地=2，林地=3，草地=4，工矿用地=5，住宅用地=6，交 通用地=7，水域=8，基塘=9，裸地=104. 在生成的具有地类属性的样本点中，再新建一个 true_class 字段，对照研究区域的spot2.5m融合图像或者google earth，在true_class字段给出真正的地类编码，具体操作 如下：4.1首先设置可用于选择的图层，仅randompoint图层可用于选择I Set Selectable LaysKChoose which lasers cari have their Jeatures selected irteractielv wit
9、h the Select Features took the Select Ely Graphics command, the Edit tool etc.回 randompoint dg dglucc4.2 对每个样本点进行编辑，给出真正的地类属性，可以通过分别对 true_class 字段值设置不同的符号化显示来区分样本点是否已经编辑过KMAti ri Lute5true_class5. 待所有的样本点编辑完毕，要进行分类误差矩阵的计算类型参考数据分 类 数 据12345678910行总 和生产者精度用户精度1X11X12X110X1+(X11/X+1)*100%(X11/X1+)*100
10、%2X21X22X210X2+(X22/X+2)*100%(X22/X2+)*100%3X31X32X33X310X3+(X33/X+3)*100%(X33/X3+)*100%4X41X44X410X4+(X44/X+4)*100%(X44/X4+)*100%5X51X55X510X5+(X55/X+5)*100%(X55/X5+)*100%6X61X66X610X6+(X66/X+6)*100%(X66/X6+)*100%7X71X77X710X7+(X77/X+7)*100%(X77/X7+)*100%8X81X88X810X8+(X88/X+8)*100%(X88/X8+)*100%9X
11、91X99X910X9+(X99/X+9)*100%(X99/X9+)*100%10X101X1010X10+(X1010/X+10)*100%(X1010/X10+)*100%列 总 和X+1X+2X+3X+4X+5X+6X+7X+8X+9X+10N概念：i为行数，j为列数生产者精度：指某一类别的正确分类数（表中主对角线上的数据Xii）占参考数据中该类别 像元总数（列数据X+i）的比例，对应的误差为漏分误差。用户精度：指某一类别的正确分类数（表中主对角线上的数据Xii）占分类数据中该类别像 元总数（列数据Xi+）的比例，对应的误差为误分误差。总体精度：指总分类正确数（表中主对角线上的数据Xii之和）占总抽样数数（N）的比例，它反映分类结果总的正确程度。 由于总体精度只利用了误差矩阵主对角线上的元素，而未利用整个误差矩阵的信息，作为分 类误差的全面衡量尚欠不足，因此可以用Kappa系数指标计算公式如下：rTK=NXii (Xi+* X + i)/N*Nr (Xi+* X + i)一=0I = 1K 为 kappa 系数；r 为分类矩阵的行数；Xii为第i行第i列的观察值；Xi+和X+i分别为分类误差矩阵的行总和及列总和；N 为总观察值Kappa 系数充分利用了分类误差矩阵的信息，可作为分类精度评估的综合指标。
