多特征组合与自动加权K-Means聚类算法的影像分类技术研究

Research on Remote Sensing Image Classification Technology Based on Multi-Feature Combining and Automatic Weighted K-Means

为了提高遥感影像地物分类精度,提出了一种基于多特征组合与自动加权K-Means聚类算法的影像分类方法。首先提取影像的SIFT,GIST,颜色,Census和Gabor等多种类型特征,然后通过实验分析确定最佳特征组合。针对一般K-Means算法没有考虑各个特征值权重的问题,提出利用自动加权K-Means算法计算不同特征分量的权值,分别对SIFT,GIST和Gabor特征构建了基于权重的影像特征词汇表;然后利用稀疏编码算法进行影像编码;最后使用SVM算法完成影像分类。通过实验表明提出的方法能有效提高遥感影像分类准确性,并且具有较好的稳定性和鲁棒性。

In order to improve the precision of remote sensing image classification, this paper proposed a new algorithm of images classification on the basis of multi features combination and automatic weighted clustering KMeans. Firstly, SIFT, GIST, Census, Gabor color and many other features are extracted from the images, then the best features combination is determined through the experimental analysis. Aim at the fact that general K-Means algorithm ignore the weight of each features, the automatic weighted K-Means algorithm was adopted to calculate the weight of different features,the images features vocabulary based on the weight for SIFT, GIST, Census and Gabor was constructed respectively. The sparse coding algorithm was adopted to code the images. Finally, the SVM algorithm was adopted to complete the images classification. The experiment results showed that the proposed method could improve the classification accuracy of remote sensing images effectively, and is more stable and robust.