2、进行分类精度评价，得到分类混淆矩阵，计算Kappa系数，并对结果进行解释。6.分类后处理（clumpsievemajority）。运用ISODATA方法进行非监督分类：预先假定地表覆盖类型为30类，迭代次数选为15，由系统完成非监督分类；然后进行类别定义与合并子类，最后进行结果的精度评价。原理和方法1、监督分类：监督分类是在分类前人们已对遥感影像样本区中的类别属性有了先验知识，进而可利用这些样本类别的特征作为依据建立和训练分类器（亦即建立判别函数），进而完成整幅影像的类型划分，将每个像元归并到相对应的一个类别中去。换句话说，监督分类就是根据地表覆盖分类体系、方案进行遥感影像的对比分析，据此建
3、立影像分类判别规则，最后完成整景影像的分类；2、可分性度量：本次实习主要涉及JM距离和变换分散度，都是一种特征空间距离度量方法，是指影像特征矢量与各个类中心的距离，变换分散度是TDivercd=1-exp(-Divercd/8),JM距离J=2*(1-e-B)；3、最大似然分类法：在两类或多类判决中，假定各类分布函数为正态分布，并选择训练区，用统计方法根据最大似然比贝叶斯判决准则法建立非线性判别函数集，计算各待分类样区的归属概率，而进行分类的一种图像分类方法。4、混淆矩阵：从随机点位上获取地面参考验证信息，并与遥感分类图进行逐像元比较，然后将结果归纳到混淆矩阵，进而完成混淆矩阵分析。类别精度：
4、被正确分类的类别像元数占该类别训练样本像元数的百分比，包括生产者精度（制图精度）和用户精度，其中制图精度对应漏分误差是指指示需要进行类别补充和训练样本的采集，用户精度对应错分误差是指指示训练样本集存在混合现象，需要进行更加精细的训练样本采集以保证各个类别样本光谱特征上的纯洁性；5、分类后处理主要/次要分析：输入一个变换核，用变换核中占主要/次要地位的像元的类别代替中心像元的类别。6、分类后处理类别集群：运用形态学算子将临近的类似分类区域合并集群。首先，将被选的类别用一个膨胀操作集群在一起，然后用参数对话框中指定了大小的变换核对分类图像进行侵蚀操作。7、分类后处理类别筛选：观察周围的4个或8
8、ldesertR215G200B18573裸地及盐碱地barrenlandR200G205B2002、按照监督分类的步骤，在影像上找出对应各个土地利用/覆盖类型的参考图斑，利用ROI工具建立训练区：训练样本如下：对训练样本进行统计，结果如下：对训练样本中各地物特征值进行统计，得到各个类别的特征统计表：地物类型73：barrenland采样单元数：波段号：1234567单变量统计最小值13329132781422116003179042056019082最大值14137143391590318212208392251620998均值13593.2613598.3014666
9、.1616604.2319011.7221217.1019863.82标准差117.51151.65254.02361.75430.88335.77305.80协方差矩阵band113807.9017388.5727130.3437109.1742342.9731679.6025318.36band217388.5722996.4636959.1751008.0256707.8141825.4433584.25band327130.3436959.1764523.7890799.1196445.3169624.1055670.50
11、0.910.870.840.800.70band20.981.000.960.930.870.820.72band30.910.961.000.990.880.820.72band40.870.930.991.000.860.790.69band50.840.870.880.861.000.890.70band60.800.820.820.790.891.000.93band70.700.720.720.690.700.931.00地物类型12：irrigatedland采样单元数：波段号：12
12、34567单变量统计最小值1025694138790793318918100797615最大值11717110821111411146330401673412689均值10689.809909.009580.438717.2323553.9912080.088899.35标准差211.99255.06471.97463.803118.971725.56890.79协方差矩阵band144938.7553378.7172959.8191906.382840.10165126.20137512.55band253378.7165057.6094072.
13、42111933.1337163.35223684.10172923.97band372959.8194072.42222751.80174786.42816795.65731138.82385861.99band491906.38111933.13174786.42215114.5542019.06415292.38325592.31band52840.1037163.35816795.6542019.069727955.684481447.711590685.73band6165126.20223684.10731138.82415
15、550.031.000.830.57band60.450.510.900.520.831.000.90band70.730.760.920.790.570.901.00地物类型30：grassland采样单元数：波段号：1234567单变量统计最小值11249106511031110437146531388111406最大值13345133411416615842208332178419584均值11879.0511460.4311642.9712181.6016989.6416400.4313878.58标准差371.87463.85
16、692.94969.591376.431535.491559.65协方差矩阵band1138285.54171492.62242267.44334994.17283385.14439751.22475810.39band2171492.62215152.57306090.13425360.86368584.97574425.20616446.05band3242267.44306090.13480159.93665417.12675080.43935368.59965996.98band4334994.17425360.86665417.
18、90.940.930.550.770.82band20.991.000.950.950.580.810.85band30.940.951.000.990.710.880.89band40.930.950.991.000.680.890.91band50.550.580.710.681.000.750.64band60.770.810.880.890.751.000.98band70.820.850.890.910.640.981.00地物类型51：stream采样单元数：波段号：1234567单变
19、量统计最小值10258938284847609683163956225最大值12215117741165311900123341090110155均值11065.2410419.3410086.719853.929014.987500.257047.20标准差478.70589.22804.611152.89996.49720.24589.85协方差矩阵band1229155.76280800.03370010.12528929.55321426.49106701.8875598.77band2280800.03347185.95463365.16
20、658887.05385331.23116783.6579117.67band3370010.12463365.16647402.47915873.44506076.14124926.5472406.60band4528929.55658887.05915873.441329147.16762081.36183308.73104583.76band5321426.49385331.23506076.14762081.36993001.35530712.68365993.41band6106701.88116783.65124926.54
22、661.000.740.62band60.310.280.220.220.741.000.97band70.270.230.150.150.620.971.00地物类型52：reservoirorpond采样单元数：波段号：1234567单变量统计最小值11103105921065310145713959805811最大值134711331313785130901092397179018均值11867.4511524.5011866.6511463.628202.856880.546583.06标准差685.33774.47840.3
23、0787.881143.201222.941054.77协方差矩阵band1469678.22530192.44571597.68488610.33775086.13815146.84704059.88band2530192.44599806.13646958.99555314.75873181.20918036.05792743.95band3571597.68646958.99706103.80615130.54933986.96982104.19848706.01band4488610.33555314.75615130.5462
25、900.990.970.97band21.001.000.990.910.990.970.97band30.990.991.000.930.970.960.96band40.900.910.931.000.870.800.80band50.990.990.970.871.000.980.98band60.970.970.960.800.981.001.00band70.970.970.960.800.981.001.00地物类型71：sandydesert采样单元数：波段号：1234567单变量统计
26、最小值12367121161278014193163251894017687最大值13574136131511017835200212384722979均值12999.5512911.5613973.2416059.4817729.4020873.4320071.00标准差184.46235.73360.60522.67724.43919.38902.80协方差矩阵band134027.0342724.4261821.9784307.15102552.99130797.22132445.86band242724.4255567.6582633.52
27、114953.12141779.18184793.23186349.87band361821.9782633.52130034.17185751.85235039.03309629.30309516.95band484307.15114953.12185751.85273188.56346740.12460868.36458951.17band5102552.99141779.18235039.03346740.12524802.69639218.08599720.02band6130797.22184793.23309629.3046
29、.921.000.960.92band60.770.850.930.960.961.000.99band70.800.880.950.970.920.991.00地物类型72：graveldesert采样单元数：波段号：1234567单变量统计最小值12536122361259013846153961707115686最大值14033141501559818228208612489222060均值13381.6113361.2014386.7716480.5519069.2121519.6919658.05标准差274.50351.45
30、545.87781.52979.351284.841152.27协方差矩阵band175350.0295950.03145101.24199282.47224200.30242453.93241388.72band295950.03123515.93188501.90260838.38296654.23324127.03320377.44band3145101.24188501.90297974.59421133.23494009.44557363.93538052.43band4199282.47260838.38421133.236
32、930.830.690.76band20.991.000.980.950.860.720.79band30.970.981.000.990.920.790.86band40.930.950.991.000.970.850.90band50.830.860.920.971.000.930.95band60.690.720.790.850.931.000.99band70.760.790.860.900.950.991.003、对所选ROI样本进行可分离性评价，结果如下：JM值统计表：Jeffries-M
33、atusita3012515271727330草地1.9991.9992.0001.9991.9981.99812水浇地1.9991.9992.0002.0002.0002.00051河流1.9991.9991.9952.0002.0002.00052水库2.0002.0001.9951.9932.0002.00071沙漠1.9992.0002.0002.0001.9991.99972砾漠1.9982.0002.0002.0001.9991.89673裸地1.9992.0002.0002.0001.99
34、91.896分离散度统计表：TransformedDivergence3012515271727330草地1.9412.0001.7852.0002.0002.00012水浇地2.0002.0002.0002.0002.0002.00051河流2.0002.0001.9802.0002.0002.00052水库2.0002.0002.0002.0002.0002.00071沙漠2.0002.0002.0002.0002.0001.99972砾漠2.0002.0002.0002.0002.0002.00073裸地2.0002.0002.0002.0002.0002.000分析上述可分性度量矩阵可知，各地物间JM值均在1.8以上，因此可以有效的对各地物进行区分，因此所选ROI样本很适合与此监督分类。4、利用最大似然法对影像数据完成监督分类。最大似然法分类后结果如下：在ENVI经典模式下，对监督分类后影像依据编码体系修改各种地物颜色，使其符合编