遥感图像识别分类技术研究

发布时间：2018-08-31 17:31

【摘要】：近年来,随着遥感技术的发展,我们能够获得极其丰富的遥感信息。但是如何处理和应用这些数据,使之转换为可以被应用的有效信息,也随之成为一个重要的问题。利用单一传统的分类方法来对遥感图像进行分类识别,不仅会导致分类精度降低,而且也会造成空间数据大量冗余以及资源浪费等问题。本文主要对遥感图像地物识别的问题进行研究。在研究过程中,通过对现有的方法进行调研与分析,采用遥感图像的不同波段信息作为特征向量,构建训练样本集与测试样本集,并选取基于多分类器组合的方法来对待测遥感图像的地物进行识别。在多分类器组合方法中,通过三层规则逐一地对待测样本进行判断,从而确保了分类结果的可靠性。接下来本文又提取出不同经纬度的20幅遥感图像进行分类识别,并与传统算法进行比较,发现整体上的分类效果要略优于传统算法。最终将生成的分类器系统与实验室自主研发的无线传输系统相结合,达到了减少通信传输量的目的。
[Abstract]:In recent years, with the development of remote sensing technology, we can obtain extremely rich remote sensing information. However, how to process and apply these data and convert them into useful information becomes an important problem. Using a single traditional classification method to classify and recognize remote sensing images will not only reduce the classification accuracy, but also lead to a lot of redundancy of spatial data and waste of resources. In this paper, the recognition of ground objects in remote sensing images is studied. In the course of the research, by investigating and analyzing the existing methods, using the different band information of remote sensing image as the feature vector, the training sample set and the test sample set are constructed. The method based on multi-classifier combination is selected to identify the ground objects of remote sensing images. In the multi-classifier combination method, three layer rules are used to judge the test samples one by one, thus ensuring the reliability of the classification results. Then 20 remote sensing images with different longitude and latitude are extracted and compared with the traditional algorithm. It is found that the overall classification effect is slightly better than the traditional algorithm. Finally, the generated classifier system is combined with the wireless transmission system developed independently in the laboratory to reduce the amount of communication transmission.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP751

【相似文献】