基于协同训练与集成学习的极化SAR图像半监督分类
发布时间：2018-10-16 18:46
【摘要】：针对全极化SAR图像在监督分类中存在的人工标注样本费时费力以及多种极化特征未能综合利用等问题,提出一种基于协同训练与集成学习的极化SAR图像半监督分类方法。该方法以支持向量机作为半监督学习的基分类器,通过协同学习机制将多种极化目标分解下的特征有效结合,实现同时利用无标注和有标注样本,最后通过集成学习进一步提高分类模型的泛化能力。在AIRSAR和EMISAR影像上的实验表明,该方法能充分利用不同特征的特点,在较少人工标注的样本下也能获得较高的分类精度。
[Abstract]:A semi-supervised classification method for polarimetric SAR images based on cooperative training and ensemble learning is proposed to solve the problems such as the time and effort of manually tagging samples in supervised classification of fully polarized SAR images and the lack of comprehensive utilization of many polarimetric features. Support vector machine (SVM) is used as the basis classifier for semi-supervised learning. By means of cooperative learning mechanism, the features decomposed by multiple polarimetric targets are effectively combined to realize the simultaneous use of unlabeled and labeled samples. Finally, the generalization ability of the classification model is further improved by integrated learning. Experiments on AIRSAR and EMISAR images show that the proposed method can make full use of the characteristics of different features and achieve high classification accuracy with fewer samples labeled manually.
【作者单位】： 河海大学地球科学与工程学院;江苏省测绘工程院;
【基金】：国家自然科学基金项目(41301449) 江苏省测绘地理信息科研项目(JSCHKY201501) 地理空间信息工程国家测绘地理信息局重点实验室经费资助项目(201324)
【分类号】：TN957.52
本文编号：2275292
[Abstract]:A semi-supervised classification method for polarimetric SAR images based on cooperative training and ensemble learning is proposed to solve the problems such as the time and effort of manually tagging samples in supervised classification of fully polarized SAR images and the lack of comprehensive utilization of many polarimetric features. Support vector machine (SVM) is used as the basis classifier for semi-supervised learning. By means of cooperative learning mechanism, the features decomposed by multiple polarimetric targets are effectively combined to realize the simultaneous use of unlabeled and labeled samples. Finally, the generalization ability of the classification model is further improved by integrated learning. Experiments on AIRSAR and EMISAR images show that the proposed method can make full use of the characteristics of different features and achieve high classification accuracy with fewer samples labeled manually.
【作者单位】： 河海大学地球科学与工程学院;江苏省测绘工程院;
【基金】：国家自然科学基金项目(41301449) 江苏省测绘地理信息科研项目(JSCHKY201501) 地理空间信息工程国家测绘地理信息局重点实验室经费资助项目(201324)
【分类号】：TN957.52
【相似文献】
相关期刊论文 前1条
1 邱诚;王大海;任伟家;邹权;;基于集成学习的音乐识别方法研究[J];计算机科学;2012年12期
相关硕士学位论文 前1条
1 雷杨;基于集成学习的垃圾短信多级分类技术研究[D];电子科技大学;2011年
，本文编号：2275292
本文链接：https://www.wllwen.com/kejilunwen/xinxigongchenglunwen/2275292.html
