遥感影像道路断线修复方法研究

在遥感影像道路提取中,由于树木遮挡或建筑物阴影等原因产生的道路断线极大影响了道路的完整性和连通性,使得其结果无法为基于路网的分析和处理方法提供可靠的数据支持。针对深度语义分割网络的遥感影像道路提取结果,提出了联合道路断线端点的位置、指向、纹理和连通性等多特征进行断点间匹配,并结合三次曲线拟合对匹配道路断点进行优化连接的道路断线修复方法。实验表明,该方法在DeepGlobe Road数据集上F1-Score相比原始结果提升了1.27%、77.04%,较大提高了原始结果的精度和可靠性。且经过断线修复后的结果具有较好的完整性和连通性,体现了该方法的对于遥感影像道路断线修复的可靠性与实用性。

Road Lateral Disconnection and Crossing Impacts in River Landscape of Lancang River Valley in Yunnan Province,China

Roads are conspicuous components in a river landscape;however,their impacts on river landscape patterns and ecological processes have not been systematically studied at the watershed scale.In this paper,the Lancang River Valley in Yunnan Province,China was selected as a case to study road lateral disconnection and crossing impacts and identify river-road network interaction.This study was primarily focused on the road impacts on soil erosion intensity and patch density by using GIS analysis at different scales and explored their distribution with terrain factors.The results showed that river density revealed spatial autocorrelation although both of the roads and rivers were distributed unevenly in the valley.The lateral road(road curvature≥1.1)proportion correlated with soil erosion intensity(p 0.01)at the small sub-basin scale.Soil erosion intensity decreased with increasing lateral road buffer width.Light erosion generally accounted for a large proportion of the erosion in the lateral road buffer zones(1.0–4.0 km),while higher class lateral roads imposed greater impacts on soil erosion than lower class roads,which primarily had a moderate erosion level.In addition,the results of road-river intersection density indicated that road crossing impacts were significantly correlated with patch density at the small sub-basin scale.Topography factor(percent of slope>25°in each sub-basin had a close relationship with the ratio of total length of road line with curvature value≥1.1 to the total number of intersections.The correlation(p 0.01)between road impacts and terrain factor revealed that topography affected the road impact distribution in the Lancang River Valley.

兼顾连通性的U-Net网络高分辨率遥感影像道路提取

遥感影像道路提取结果中的断线一方面降低了提取精度,另一方面影响了道路形态完整性,使得提取结果不能直接应用于空间决策与分析。本文基于U-Net网络在高分辨率遥感影像道路提取时全局特征表达的优势,提出一种兼顾连通性的道路断线修复方法完善U-Net网络局部特征表达的劣势。首先,利用数据增强和扩充数据量后的样本数据作为U-Net网络的输入以此训练模型并进行最优模型的道路提取;然后,对提取结果中出现的道路断线以三次多项式曲线拟合的形式进行优化处理。实验表明,与相近网络比较,本文道路提取的精度和形态完整性有了明显的提高,查准率为86.25%,查全率为85.50%,F1-score达到了85.87%。其成果数据能直接地应用于地理决策分析,特别有利于灾后的路径规划,本文提出的方法对道路、电网、轨道、河流等线性地物分类结果中出现类似断线问题具有一定的参考意义。