目的 基于中高分辨率影像进行大范围的农村建筑区提取时,由于影像分辨率的限制以及农村建筑区规划自身规划特点等因素,造成了传统变差函数方法的高错分误差。为了准确提取农村建筑区,为后续获取建筑密度和人口密度等工作建立基础,提出了一种基于迭代P参数法的阈值确定方法。方法 通过设定亮度阈值,在变差函数纹理计算中为满足条件的像元点赋以权值。本文方法确保在4个方向都满足条件的像元点(认为是建筑区)获得较大的变差函数值加成,而仅在一个方向或者没有方向满足条件的像元点(认为是非建筑区)获得较小加成或不变,以此改进传统变差函数方法,抑制了农村建筑区与周边非建筑区的混淆。结果 以Radarsat-2的多个极化波段影像为数据源进行了实验,改进变差函数方法在实验区1与实验区2的各个波段平均检测率分别为91.58%和90.11%,平均错分误差分别为19.83%和31.87%。结论 与传统变差函数方法以及最小距离法相比,既保证了较高的检测率,同时显著降低了错分误差,不足之处是在建筑区与非建筑区的边缘处以及与建筑区具有相似纹理特征的非建筑区处出现错分,需要进一步的研究和完善。

Objective With rapid development of the social economy and the urbanization process, building land has increased continuously, promoting the use of remote sensing images on built-up area extraction, which lays the foundation for detecting buildings and obtaining indices, such as building density and population density. Variogram is proven to be an effective methodology in extracting texture characterization with high precision when applied to urban built-up area extraction from high-resolution synthetic aperture radar (SAR) images. However, limited by factors, such as imagery resolution and structural features of rural built-up areas, the traditional way of executing the variogram is characterized by a high false alarm rate when extracting rural built-up areas in a large extent from middle- and high-resolution SAR images. Aimed at extracting rural built-up areas precisely, ensuring a high detection rate, and lowering the false alarm rate, this study proposed a new method to determine the threshold based on the iterative P-parameter. Method Through the establishment of the brightness threshold, pixels that meet the requirement can be assigned a weight to increase the value of the variogram for pixels up to a standard in four directions (which can be considered built-up areas) and to restrain the increase of pixels only fitting one direction or having no direction (which can be considered non-built-up areas). Thus, the confusion between built-up areas and its counterpart is reduced. The experiment employed Radarsat-2 image as data source. According to the features of the variogram, cross-polarization images, VH and HV bands, and their principal component analysis transformation, which are characterized by a high discrepancy between built-up areas and non-built-up areas, are selected as input data. The first step is to determine the threshold and weight based on the iterative P-parameter. The second step is to extract the feature vectors of the image using the variogram. The third step is binarization with the FCM classifier. The final step is to determine whether to recalculate or not according to the extraction precision. Results The results indicate that the mean detection rates of the advanced variogram method on experimental areas 1 and 2 are 91.58% and 9011%, respectively, whereas the mean false alarm rates are 1983% and 3187%, respectively. Compared with the traditional and minimum distance methods, the proposed method not only maintains a relatively high detection rate but also decreases the false alarm rate significantly. Conclusion The precision of rural built-up area extraction basically satisfies the requirements of practical application. However, the method still has several drawbacks. First, the "edge effect", which refers to the phenomenon that the edge of built-up areas extracted from the image are actually wider than the actual edge, accompany almost every part of the extracted built-up areas. Meanwhile, this methodology is unable to recognize regions sharing similar texture features with built-up areas, such as the highlighted thready road and lumpish arable land, leading to false recognition of built-up areas. To solve the aforementioned problems, a method to restrain the high value of the variogram in places adjacent to the boundary of building areas and preprocessing are needed to ensure that certain texture features symbolize only the built-up areas.