论文部分内容阅读
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。
The traditional dictionary-based visual dictionary ignores the category information contained in the scene itself and makes it difficult to distinguish between different categories but with similar appearance. To solve this problem, this paper proposes a visual word optimization method that takes into account the category of scene information, and uses Boiman’s distribution strategy And principal component analysis to optimize ambiguity and word redundancy of visual words in different scene categories and enhance the visual dictionary’s recognition ability. By calculating the image frequency of different visual words, the algorithm removes the visual words in the visual dictionary, which have a lesser frequency of images, obtains a category visual dictionary for each scene, computes the category histogram, and merges the category histogram and the original visual histogram to obtain the different The fusion histogram of the category scenes is trained and classified as the input vector of the SVM classifier. The experimental results show that this algorithm can adapt to different size of the visual dictionary, increase the scene category information in the model, enhance the recognition ability of the phrase package model, effectively reduce the scene misclassification probability, the overall classification The accuracy is up to 89.5%, which is better than the traditional remote sensing image scene classification algorithm based on pyramid matching word packet model.