一种利用空间和光谱信息的高光谱遥感多分类器动态集成算法

发布时间：2018-06-09 03:24

本文选题：高光谱遥感 + 多分类器动态集成　；参考：《国土资源遥感》2017年02期

【摘要】：针对高光谱遥感影像分类面临的小样本、分类器不稳定等问题,在总结现有多分类器动态集成算法的基础上,提出了一种利用空间和光谱信息的多分类器动态集成算法。首先,采用支持向量机等5个基分类器构建多分类器集合;其次,计算各个分类器的分类结果,将大多数分类器分类一致的像元列入样本数据;最后,根据待分类像元的邻域像元的标签分类情况,动态地选择合适的方式进行分类器集成。该算法只在空间邻域信息满足一定条件的情况下,才采用空间和光谱信息结合的方法进行处理,即利用空间信息提高算法的灵活性。采用2幅不同传感器的高光谱遥感影像数据对算法进行实验,并与现有5种多分类器动态集成算法进行对比分析。结果表明,本文提出的多分类器动态集成算法可以保持较高的分类精度,并能有效提升高光谱遥感影像分类的稳定性,对于推动高光谱遥感精细分类研究具有一定的理论和实用价值。
[Abstract]:Aiming at the problems of small sample and unstable classifier in hyperspectral remote sensing image classification, a dynamic ensemble algorithm based on spatial and spectral information is proposed on the basis of summarizing the existing multi-classifier dynamic ensemble algorithms. First of all, support vector machine and other five basis classifiers are used to construct multi-classifier sets. Secondly, the classification results of each classifier are calculated, and most of the pixels that are consistent in classifier classification are included in the sample data. According to the label classification of neighboring pixels of the pixel to be classified, a suitable method is dynamically selected for classifier integration. The method of combining spatial and spectral information is used to improve the flexibility of the algorithm only if the spatial neighborhood information meets certain conditions. Two hyperspectral remote sensing image data of different sensors are used to test the algorithm, and compared with five existing multi-classifier dynamic ensemble algorithms. The results show that the proposed multi-classifier dynamic ensemble algorithm can maintain high classification accuracy and effectively improve the stability of hyperspectral remote sensing image classification. It has certain theoretical and practical value for promoting the research of hyperspectral remote sensing fine classification.
【作者单位】：河海大学地球科学与工程学院;武汉大学测绘遥感信息工程国家重点实验室;
【基金】：国家自然科学基金项目“高光谱遥感影像多特征优化模型与协同表示分类”(编号:41571325);“基于共形几何代数的高光谱遥感影像降维与分类”(编号:41201341)共同资助
【分类号】：P237

【相似文献】