论文部分内容阅读
通过对自动化样本选择方法进行研究,实现了局部区域内面向对象的土地覆被自动分类。首先通过模糊聚类获得影像中的候选对象样本,分别提取影像特征和先验知识中的地类特征,通过预设阈值完成样本初步筛选,然后根据先验知识进行半监督距离度量学习,完成样本的自动选择,并为最终的监督分类提供度量依据。应用舟曲泥石流灾区影像进行了实验,结果表明,本文方法与基于人工选择样本的分类结果精度非常接近,同时在多次实验中表现出较高的稳定性,相对人工方法更加客观,适合批量自动化处理。
Through the study of the method of automatic sample selection, the object-oriented automatic land cover classification in the local area is realized. Firstly, the candidate samples in the image are obtained by fuzzy clustering, and the features of the geospatial features in the image features and the prior knowledge are respectively extracted. The preliminary screening of the samples is completed by the preset thresholds. Then the semi-supervised distance metric learning based on the prior knowledge is completed, Automatic selection, and to provide the basis for the final classification of supervision. The experiment results show that the accuracy of the proposed method is close to the accuracy of the classification results based on the artificial selection samples, meanwhile it shows high stability in many experiments. Compared with artificial methods, the method is more objective and suitable for batch automation deal with.