结合显著性检测和超像素分割的遥感信息提取算法研究

Extraction algorithm of remote sensing information based on saliency detection and superpixel segmentation

针对目前显著性检测算法在复杂多目标遥感图像中检测能力不足的问题,提出一种结合显著性检测和超像素分割的遥感信息提取算法。该算法通过GBVS(graph-based visual saliency)方法检测出原始影像中部分显著性较高的区域,然后利用SLIC(simple linear iterative clustering)方法分割显著区域,并修正显著区域边缘得到训练样本数据,进一步对训练样本进行统计学习,构造显著目标提取的阈值区间,最后实现对整幅超像素图像的显著目标提取。实验结果表明,该算法具有较高的准确率和召回率,能更加有效地检测出遥感图像中的显著目标,比目前主流的显著区域检测算法提取效果更好,可以很好地应用于具有明显显著区域的复杂多目标遥感图像信息提取中。

Due to the weakness of complex multi-target detection using some saliency detection algorithms in remote sensing images,this paper proposed an extraction algorithm based on saliency detection and superpixel segmentation. Firstly,it used graph-based visual saliency（ GBVS） method to extract some high saliency regions correctly,then applied simple linear iterative clustering（ SLIC） method to segment these saliency regions into some superpixels,and meanwhile amended part of the edge superpixels,and as the training sample. Secondly,by computing the statistical feature parameters of the training sample and constructing a reasonable statistic,it established a saliency objects extraction threshold range. Finally,it used the threshold range to extract the salient objects from the whole superpixels successfully. Compared with other saliency regions detection methods,experimental results show that the proposed algorithm has more higher precision and recall values and can detect the saliency objects more efficient,and can be better applied to extract remote sensing saliency information from the complex multitarget remote sensing images.