高分辨率遥感图像中建筑物提取的边界精细化卷积神经网络

A Boundary-Refined Convolutional Neural Network for Building Extractionin High-Resolution Remote Sensing Imagery

随着遥感图像解译在城乡规划和数字化城市建设等领域的广泛应用,对遥感图像中的建筑物进行完整地、准确地检测具有非常重要的研究意义和应用价值。近年来,深度学习技术被广泛应用于遥感图像中的建筑物提取。然而,如何从高分辨率遥感图像中完全地、准确地提取建筑物仍然面临着巨大的挑战。因此,文章提出了一种边界精细化的建筑物提取方法,命名为BR-Mask R-CNN。首先,文章采用多特征融合网络ResNeXt101-FPN作为主干特征提取网络,以提高小型建筑物的提取精度。然后,利用边界精细化掩码分支将边界保护分支和Mask分支集成起来,以保护建筑物的边界信息,并实现更加准确的掩码预测。最后,文章在两个公开的建筑物提取数据集上验证了所提出方法的有效性,实验结果表明该文的方法在许多评价指标上都有较好的效果。

With the widespread application of remote sensing image interpretation in the fields of urban and rural planning and digital city construction,the complete and accurate extraction of buildings in remote sensing images has very important research significance and application value.In recent years,deep learning technology has been widely used in building extraction from remote sensing images.However,how to completely and accurately extract buildings from high-resolution remote sensing images stil faces huge challenges.Therefore,the paper proposes a boundary-refined building extraction method called BR-Mask R-CNN.Firstly,the paper adopt the multi-feature fusion network ResNeXt101-FPN as feature extraction network to improve the extraction accuracy of small buildings.Then,the boundary-refined mask branch is utilized to integrate the boundary-preserving branch and the mask branch to protect the boundary information of the buildings,and to achieve more accurate mask prediction.Finally,the paper employ two public building extraction datasets to verify the effectiveness of the proposed method,and experimental results demonstrate that my approachhas better results on many evaluation metrics.