论文部分内容阅读
目的现有基于结构分析的高分辨率SAR影像建筑物检测方法,只考虑了直线和L形结构建筑物,并且依赖建筑物高亮线条处阴影区作为建筑物识别的主要特征;当处于复杂场景时,阴影区受制于背景较暗或建筑物密集而无法准确得到,导致建筑物检测误差大、检测率低。针对上述问题,提出一种基于形态学层级分析的高分辨率SAR影像无监督建筑物检测算法。方法该方法基于单幅单极化高分辨率SAR影像,首先利用改进的形态学交替滤波算子有效抑制其固有的斑点噪声,大大剔除了同质区背景噪声的干扰;然后利用层级分析形态学差分属性断面算法来实现对SAR影像建筑物的几何结构特征的提取;最后结合特征融合和属性阈值分割等后处理步骤得到复杂场景下建筑物提取信息。结果将上述方法在建筑物密集的城区SAR影像中实验,通过与其他方法对比分析,具有检测率高、误差小的特点,准确率和召回率分别为95.38%、86.31%,并对降低虚警率方面有明显的优势。结论将形态学交替滤波与形态学属性滤波的改进与结合,在对不同走向、尺寸和形状的高密度建筑物检测中具有较好的适应性。
The existing structural detection based on high resolution SAR image buildings detection method, considering only the linear and L-shaped structure of buildings, and relies on the shaded area of the building highlighted as the main features of building identification; when in complex scenes , The shaded area is subject to the darker background or the dense building and can not be accurately obtained, resulting in large building detection error and low detection rate. In view of the above problems, this paper proposes an unsupervised detection algorithm of high-resolution SAR images based on morphological hierarchy analysis. Methods Based on a single-polarimetric high-resolution SAR image, this method firstly uses an improved morphological alternating filter to effectively suppress speckle noise and greatly eliminates the interference of background noise in the homogenous region. Then, Differential attribute section algorithm to extract the geometric structure features of SAR image buildings. Finally, combined with feature fusion and attribute threshold segmentation and other post-processing steps to get the complex scene under building information extraction. Results The above method was tested in SAR images of densely populated urban area. Compared with other methods, the method has the advantages of high detection rate and small error, the accuracy rate and recall rate were 95.38% and 86.31%, respectively. Rate has obvious advantages. Conclusion The combination of morphology alternation filtering and morphological property filtering has good adaptability in detecting high-density buildings of different directions, sizes and shapes.