基于SIFT和非参贝叶斯的高分辨率遥感影像地物识别算法

摘要

Object recognition has always been a significant problem in the field of remote sensing image processing.With the development of remote sensing technology,the high resolution remote sensing images carry plenty of scale invariant features which are highly correlated with each other,and the traditional recognition methods are difficult to adapt this development.Based on the SIFT(Scale-invariant Feature Transform)algorithm,a fast and accurate algorithm for ground object recognition was proposed,namely DBSIFT(Double Backward SIFT).This method constructs the new pyramid based on SIFT,uses DP(Dirichlet Process) to identify the similar features,and then segments them.Weighing upon the relationship between geometry and arithmetic,nine indexes were selected to evaluate the accuracy of segmentation.In the experiment,similar ground object can be identified accurately,and the segmentation result's performance is well.The effectiveness of this method is further explained.%地物识别是遥感图像处理领域中的一个重要问题.随着遥感技术的发展,高分辨率遥感影像中携带有大量相似的具有尺度不变特征的地物,传统的地物识别方法难以适应这一发展,亟需对其进行改进.针对高分遥感影像,在SIFT(Scale-invariant Feature Transform)算法的基础上进行改进并得出一种快速精准的地物识别算法DBSIFT(Double Backward SIFT),实现了相似地物多对一的模式识别.DBSIFT在原算法的基础上构造了二重差金字塔,利用DP(Dirichlet Process)识别出相似地物并对其进行分割.在几何与算数关系上,选取9个指标对分割精度进行评价.实验中,使用该方法得到的地物能够被准确识别,且分割效果良好,说明了该算法的有效性.