顾及遥感影像场景类别信息的视觉单词优化分类

发布时间：2018-03-11 02:14

本文选题：场景类别　切入点：类别直方图　出处：《遥感学报》2017年02期 　论文类型：期刊论文

【摘要】：传统词包模型的视觉词典忽略了场景本身包含的类别信息,难以区分不同类别但外观相似的场景,针对这个问题,本文提出一种顾及场景类别信息的视觉单词优化方法,分别使用Boiman的分配策略和主成分分析对不同场景类别视觉单词的模糊性和单词冗余进行优化,增强视觉词典的辨识能力。本文算法通过计算不同视觉单词的影像频率,剔除视觉词典中影像频率较小的视觉单词,得到每种场景的类别视觉词典,计算类别直方图,将类别直方图和原始视觉直方图融合,得到不同类别场景的融合直方图,将其作为SVM分类器的输入向量进行训练和分类。选取遥感场景标准数据集,验证算法,实验结果表明:本算法能适应不同大小的视觉词典,在模型中增加场景类别信息,增强了词包模型的辨识能力,有效降低场景错分概率,总体分类精度高达89.5%,优于传统的基于金字塔匹配词包模型的遥感影像场景分类算法。
[Abstract]:The visual dictionary of traditional lexical packet model ignores the category information contained in the scene itself, and it is difficult to distinguish the scene with different categories but similar appearance. In view of this problem, this paper proposes a visual word optimization method which takes into account the scene category information. The fuzzy and redundancy of visual words in different scene categories are optimized by using Boiman's assignment strategy and principal component analysis, and the recognition ability of visual dictionaries is enhanced. This algorithm calculates the image frequency of different visual words. By eliminating the visual words with less image frequency in the visual dictionary, the category visual dictionary of each scene is obtained, the category histogram is calculated, the category histogram and the original visual histogram are fused, and the fusion histogram of different kinds of scene is obtained. It is used as input vector of SVM classifier for training and classification. The standard data set of remote sensing scene is selected to verify the algorithm. The experimental results show that the algorithm can adapt to different size visual dictionaries and add scene category information to the model. The recognition ability of the word packet model is enhanced and the probability of scene misclassification is effectively reduced. The overall classification accuracy is as high as 89.5, which is superior to the traditional remote sensing image scene classification algorithm based on pyramid matching lexical packet model.
【作者单位】：武汉大学测绘学院;
【基金】：国土资源部公益性行业科研专项(编号:201511009-01)
【分类号】：TP751

【相似文献】