摘要
针对标签传播算法缺乏对新生成样本的评价进而影响分类精度的问题,本文提出一种利用阈值的标签传播算法来提高高光谱图像的分类精度。首先,用基于图像融合和递归滤波的特征提取方法对原始高光谱图像进行处理。然后,给出一个阈值并对标签传播算法新生成样本进行评价,保留一些可信度较高的样本。最后,保留的新样本和已标记样本之和作为训练样本,对图像进行分类。实验表明,基于改进标签传播算法优于其他的高光谱图像分类算法。
To solve the lack of evaluation of the label propagation algorithm for new samples which further affects the classification accuracy, this paper proposed a new label propagation algorithm about the threshold to improve the classification accuracy of hyperspectral images.First of all, the original hyperspectral images were processed with the method of feature extraction based on image fusion and recursive filtering.Then a threshold was given and the new samples produced by label propagation algorithm were evaluated.Some samples with higher credibility were kept.Finally, with the newly-kept samples and tagged samples as the training samples, the images were classified.Experimental results show that the modified label propagation algorithm is better than other hyperspectral image classification algorithms.
出处
《山东科技大学学报（自然科学版）》
CAS
2016年第6期101-107,共7页
Journal of Shandong University of Science and Technology(Natural Science)
基金
国家自然科学基金青年基金项目(41406200)
山东省自然科学基金青年基金项目(ZR2014DQ030)
关键词
标签传播
高光谱图像分类
阈值法
递归滤波
label propagation
hyperspectral image
threshold value method
recursive filtering
