兼顾连通性的U-Net网络高分辨率遥感影像道路提取

Road extraction of high-resolution satellite remote sensing images in U-Net network with consideration of connectivity

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。

Existing remote sensing image interpretation methods can obtain high-precision classification results but also have some defects.On the one hand,the problem of disconnection in the road extraction result of remote sensing image reduces the extraction precision.On the other hand,this problem affects the integrity of the road morphology.Thus,the extraction result cannot be directly applied to spatial decision making and analysis.Therefore,this study solves the problem of disconnection in the road extraction results of remote sensing images.A road extraction method for high-resolution remote sensing image of U-Net network with consideration of connectivity is also proposed.On the basis of the advantages of global feature representation of U-Net network in road extraction of high-resolution remote sensing images,this study proposes a broken road repair method that considers connectivity to improve the local features of U-Net network.First,the sample data after data enhancement and data volume expansion are used as an input of the U-Net network to train the model and perform road extraction of the optimal model.Then,the road breakpoint detection,the road breakpoint clustering,and the cubic polynomial curve fitting are organically combined to optimize the result because of the broken road appearing in the extraction result.The proposed method is practical and feasible according to the verification of experimental results.The method is also universal.The experiment shows that the accuracy and shape integrity of the road extraction in this method are improved significantly compared with similar networks.The precision is 86.25%,the recall rate is 85.50%,and F1-score reaches 85.87%.The roads extracted by the proposed method have good connectivity and morphological integrity.The resulting image can be directly applied to geographic decision analysis,especially for post-disaster path planning.The proposed method has certain reference significance for the occurrence of similar disconnection problems in the classification results of linear objects,such as roads,power grids,orbits,and rivers.