论文部分内容阅读
将地学知识与影像标定相结合,一直是目视解译或计算机自动分类制图的主要手段。传统的目视解译方法能够充分利用地学知识,但需要大量的人力、物力,效率较低;计算机分类中尚未出现比较成熟的高效运用地学知识的分类方法。已有研究表明,分类样本可以作为地学知识的载体,将地学知识融入分类过程中;此外,无监督聚类可以显著提高样本选取的效率,有助于提供足够的样本,为将地学知识高效地融入计算机分类提供了一定的基础。本文提出一种以前期土地利用数据辅助与影像聚类相结合的样本自动选取方法。利用自动选取的样本,通过最大似然分类器对TM影像进行分类,并与手动选取样本分类的方法进行了对比分析。研究结果表明,在分类效果上,本文提出的前期土地覆被辅助下的分类样本自动选取方法,优于手动选取样本的方法,提高了分类效率。在水体、林地、园地、城镇建设用地等7种类型上的分类整体精度达到84.18%,kappa系数为0.8066;手动选取样本进行分类的整体精度为77.04%,kappa系数为0.7196。
The combination of geo-knowledge and image calibration has been the main means of visual interpretation or automatic classification of computer graphics. The traditional method of visual interpretation can make full use of geo-knowledge, but requires a lot of manpower, material resources, low efficiency; computer classification has not yet appeared more efficient and efficient use of geo-knowledge classification. It has been shown that classified samples can be used as a carrier of geo-knowledge to integrate geo-knowledge into the classification process. In addition, unsupervised clustering can significantly improve the efficiency of sample selection and help to provide enough samples. In order to effectively integrate geo-knowledge Into the computer classification provides a certain basis. This paper presents a method of automatic sample selection based on the combination of previous land use data and image clustering. Using the automatically selected samples, the TM images are classified by the maximum likelihood classifier and compared with the method of manually selecting the sample classification. The results show that in the classification effect, the method of automatic selection of classification samples aided by the earlier land cover presented in this paper is superior to the method of manually selecting samples, which improves the classification efficiency. The overall accuracy of the classification of water body, forest land, garden land and urban construction land reached 84.18% and the kappa coefficient was 0.8066. The overall accuracy of manual classification of the samples was 77.04% and the kappa coefficient was 0.7196.