摘要
分析了当遥感分类类别存在光谱特征重叠时,以训练区数据估计类别总体特征发生偏差的原因和性质,提出运用众数滤波(Majority滤波)对训练区进行纯化.一个例子的研究表明,尤其在提取某一类或少数几类专题信息时,如果不关心其余专题信息的提取精度,Majority滤波是一种较好训练区纯化方式.
This paper analyses the attribution and cause of the deviation of estimating the class population's feature by training samples.We perform a supervised classification on the remote sensing imagine and some classes have overlapping on their distribution curves,and propose using the technology of mode filtering (Majority filtering) to carry on the purification of the training area.One example shows,especially at the time of drawing some one or a few several kinds of thematic information.If the accuracy of abstraction of other thematic information is not considered,majority filtering is a better purification way of training area.
出处
《信阳师范学院学报（自然科学版）》
CAS
2003年第3期309-313,共5页
Journal of Xinyang Normal University(Natural Science Edition)
基金
河南省优秀骨干教师资助项目(2001-12)
关键词
训练样本
纯化
监督分类
training samples
purification
supervised classification
