Lab6 non-parametric classification and post classification
12021005龚鑫烨Objection:the major object of the current lab section are to implement non-parametric classification based on BP networks and support vector machines algorithms,with a full mastery of post-classification operation. Data: the subset of spot 5 imagery covering NJ.
Steps:
1、identify a training dataset and an independent set of validation data for built-up, forest,cropland,grassland and water.
2、Implementing above-mentioned non-parametric algorithms to classify your image.
3、Validating your classification.
4、Refining your classification by implementing the majority filtering and modeling process if possible.
实验步骤:
1、将数据加载到envi中
building、water、grass)
保存ROI
3、BP分类。
Classification——supervised——neural net,设置参数及输出路径
观察RMS动态
加载变换后的图像,和原图像进行对比
Classification——post classification——confusion matrix——using ground truth ROIs
由上图可以看出精确度为99.8%
Bp分类的校正
Classification——post classification——majority analysts ,进行网格设置
通过对这两个图层地理连接,查看校正的效果
4、svm分类方法:classification——supervised——support vector machine
Svm分类效果的验证
Classification——post classification——confusion matrix——using ground truth ROIs
Svm分类的校正
Classification——post classification——majority analysts ,进行网格设置
将生成的图像与svm图像进行地理连接,查看校正效果
Basic tools——sunset data via ROIs
5、erdas里建模修改误分的像元
以support vector machine分类的图像为例
将切好的图像和之前的svm图像加载进来,并修改他们的投影信息为基于WGS84的UTM 投影。
Molder-----molder maker,建立模型
建立关系式
运行模型后加载图像
(模型一运行就闪退。
)最后进行GEO link地理连接。