资源描述
老师要求提交:1. 可能性矩阵2. 精度评价报告3. 分类结果图具体流程:QuickBird影像为例:为真彩色,即地物的真实颜1. 打开影像,考试时的影像是老师给的高分辨率影像。以已有的File-Ope n Image File,在 Available Ba nd 中以 RGB 打开, 色。Gray* RGB Cel or金 r R:O52691247010_01_P00L tif,G G:O52691247010_01_P00k tiC J0:052691247010JlPOOl. tifLoad RGB Display #1 *2选择监督分类样本(感兴趣区域):在影像的工具栏中选择,Overlay-Region of in terest在打开的#1 ROI Tool工具栏中,以多边形的方式选择感兴趣区:在zoomROI-TypePolygon窗中进行选择选择类别,植被,水体,裸地,房屋。Fil* E0I_Typ & Options HelpKOI NameColorPixelsF olygons*水体.Blue2,2165/2,216 a探地Yellow2, 2594/2,259房屋Red2,5528/2,552植被Gre曲2,9344/2.934 IAK R芒胃ionStatsGrewfisel | Dleti n: 宀C Scroll Zoom r OffSelect AllHide ROIs how ROIs查看分离程度,继续在 ROI Tool工具栏中,选择Option compute ROI separability, 选择影像ok.Separation (least tu most):Yellow 22S9 points and 房屋Red 2552 points - 1.77012901 Blue 2216 points nd 推破Green 2934 poin ts - 1 .八2308515Blue 2216 points and 房屋Ked 25S2 points - 1 9276451 ; Red 2552 口 cints and 植械Green 2934 points - 1,99999148 Yellow 22E9 points and 植被Green 2 934 points - 2. OOOOOOOO ; Biue 2216 points and 襦地Yellow 2259 points - 2 .00000000相关度大于1.8的说明分类较好。保存文件。2. 用最大似然法进行监督分类,主菜单栏中,Classificatio n Supervised MaximumLikelihood,进入选择参数的对话框。Select all Item阈值 Probability Threshold 一般在 01 之间。不需输出真实值。因为还要分类后处理,储存至memory.ChHpwt Result FileOutput Rule Images;卞单 laiziMUA Likelihood ParaAeters水萍Blue 2216 points Ml Tellow 2259 points 房屋Bed! 2552 points 植镀Gr 电皂 n 0M34 point 百Hwfibtr items seltcltd: pSelect All Ttems Clear All ItemsSet Frolbability ThresholdHnt r* Sin (lt Multi.pl* VtlussProbability ThrmEhciLdL3分类后处理,分类合并,在主菜单中n?5Classificati on 一 post classificati on 一 Sieve ClassesF11 fe Information:052691247010 01 FOOl. tifFile: LNIemory3Dims: 8452 x10T40 JC 1 ESQSiz: LByts File90, 7T4, 480 Lytes EHVT Type : SensorClusifi c&ti onTypa: Byte OrdeVRJCTICWA Projection:Hast dntfel)tfaivaltiigtli : UppKoneLeft Dcseriptiftik:ane Classi icti 12:47omer: 1,12013Nueimum Likelihood onTu Mar 19Spatial Subset FxHl Scene0KC aitC elOp e-n 选择刚才分类好的,memory影像,改变Group Min Threshold数值,由2改到8即改变每类 别最小像元值,由于我只选了四个类别数,应该做完后不会有类别的合并。保存文件,即要求交的分类结果图。生成混淆矩阵主菜单中,Classificatio n一 post classificatio n 一 con fusi on Matrix 一 Using Ground TruthROIS.与 Iatcih Classes ParasetersSelectGround Truth RDISelect Classi ficBti on ImageGround Truth ROI裸地CLissification CLAssAdd CombinationMatched 01 assesBlue 2216 pointsOK Cancel保存混淆矩阵A驴 Class Confusion MatrizFileConfusion Matrix: Memory3 (8452x10740x1)98.2231XOverall Accuracy = (9794/9961)Ka 口口 a Coef f icient = 0.9763ClassGround Truth 水体房屋UruclaiSKi f ied632 6水怖Blue 2蒔地2209105Yellow房垦Red240925 植被 Green 1 051_o70Total2216小y2552
展开阅读全文