Building segmentation from high-resolution synthetic aperture radar (SAR) images has always been one of the important research issues. Due to the existence of speckle noise and multipath effect, the pixel values chang...Building segmentation from high-resolution synthetic aperture radar (SAR) images has always been one of the important research issues. Due to the existence of speckle noise and multipath effect, the pixel values change drastically, causing the large intensity differences in pixels of building areas. Moreover, the geometric structure of buildings can cause strong scattering spots, which brings difficulties to the segmentation and extraction of buildings. To solve of these problems, this paper presents a coherence-coefficient-based Markov random field (CCMRF) approach for building segmentation from high-resolution SAR images. The method introduces the coherence coefficient of interferometric synthetic aperture radar (InSAR) into the neighborhood energy based on traditional Markov random field (MRF), which makes interferometric and spatial contextual information more fully used in SAR image segmentation. According to the Hammersley-Clifford theorem, the problem of maximum a posteriori (MAP) for image segmentation is transformed into the solution of minimizing the sum of likelihood energy and neighborhood energy. Finally, the iterative condition model (ICM) is used to find the optimal solution. The experimental results demonstrate that the proposed method can segment SAR building effectively and obtain more accurate results than the traditional MRF method and K-means clustering.展开更多
To address the current issues of inaccurate segmentation and the limited applicability of segmentation methods for building facades in point clouds, we propose a facade segmentation algorithm based on optimal dual-sca...To address the current issues of inaccurate segmentation and the limited applicability of segmentation methods for building facades in point clouds, we propose a facade segmentation algorithm based on optimal dual-scale feature descriptors. First, we select the optimal dual-scale descriptors from a range of feature descriptors. Next, we segment the facade according to the threshold value of the chosen optimal dual-scale descriptors. Finally, we use RANSAC (Random Sample Consensus) to fit the segmented surface and optimize the fitting result. Experimental results show that, compared to commonly used facade segmentation algorithms, the proposed method yields more accurate segmentation results, providing a robust data foundation for subsequent 3D model reconstruction of buildings.展开更多
This paper studies the effect of breaking single-class building data into multi-class building data for semantic segmentation under end-to-end architecture such as UNet, UNet++, DeepLabV3, and DeepLabv3+. Although, th...This paper studies the effect of breaking single-class building data into multi-class building data for semantic segmentation under end-to-end architecture such as UNet, UNet++, DeepLabV3, and DeepLabv3+. Although, the already existing semantic segmentation methods for building detection work on the imagery of developed world, where the buildings are highly structured and there is a clearly distinguishable space present between the building instances, the same methods do not work as effectively on the developing world where there is often no clear differentiable spaces between instances of building thus reducing the number of detected instances. Hence as a noble approach, we have added building contours as new class along with building segmentation data, and detected the building contours and the inner building regions, hence giving the precise number of buildings existing in the input imagery especially in the convoluted areas where the boundary between the buildings are often hard to determine even for human eyes. Breaking down the building data into multi-class data increased the building detection precision and recall. This is useful in building detection where building instances are convoluted and are difficult for bare instance segmentation to detect all the instances.展开更多
Building extraction from remote sensing data is an important topic in urban studies and the deep learning methods have an increasing role due to their minimal requirements in training data to reach outstanding perform...Building extraction from remote sensing data is an important topic in urban studies and the deep learning methods have an increasing role due to their minimal requirements in training data to reach outstanding performance.We aimed to investigate the original U-Net architecture’s efficiency in building segmentation with different number of training images and the role of data augmentation based on multisource remote sensing data with varying spatial and spectral resolutions(WorldView-2[WV2],WorldView-3[WV3]images and an aerial orthophoto[ORTHO]).When the trainings and predictions were conducted on the same image,U-Net provided good results with very few training images(validation accuracies:94-97%;192 images).Combining the ORTHO’s and WV2’s training data for prediction on WV3 provided poor results with low F1-score(0.184).However,the inclusion of only 48 WV3 training images significantly improved the F1-score(0.693),thus,most buildings were correctly identified.Accordingly,using only independent reference data(other than the target image)is not enough to train an accurate model.In our case,the reference from WW2 and ORTHO images did not provide an acceptable basis to train a good model,but a minimal number of training images from the targeted WV3 improved the accuracy(F1-score:69%).展开更多
文摘Building segmentation from high-resolution synthetic aperture radar (SAR) images has always been one of the important research issues. Due to the existence of speckle noise and multipath effect, the pixel values change drastically, causing the large intensity differences in pixels of building areas. Moreover, the geometric structure of buildings can cause strong scattering spots, which brings difficulties to the segmentation and extraction of buildings. To solve of these problems, this paper presents a coherence-coefficient-based Markov random field (CCMRF) approach for building segmentation from high-resolution SAR images. The method introduces the coherence coefficient of interferometric synthetic aperture radar (InSAR) into the neighborhood energy based on traditional Markov random field (MRF), which makes interferometric and spatial contextual information more fully used in SAR image segmentation. According to the Hammersley-Clifford theorem, the problem of maximum a posteriori (MAP) for image segmentation is transformed into the solution of minimizing the sum of likelihood energy and neighborhood energy. Finally, the iterative condition model (ICM) is used to find the optimal solution. The experimental results demonstrate that the proposed method can segment SAR building effectively and obtain more accurate results than the traditional MRF method and K-means clustering.
文摘To address the current issues of inaccurate segmentation and the limited applicability of segmentation methods for building facades in point clouds, we propose a facade segmentation algorithm based on optimal dual-scale feature descriptors. First, we select the optimal dual-scale descriptors from a range of feature descriptors. Next, we segment the facade according to the threshold value of the chosen optimal dual-scale descriptors. Finally, we use RANSAC (Random Sample Consensus) to fit the segmented surface and optimize the fitting result. Experimental results show that, compared to commonly used facade segmentation algorithms, the proposed method yields more accurate segmentation results, providing a robust data foundation for subsequent 3D model reconstruction of buildings.
文摘This paper studies the effect of breaking single-class building data into multi-class building data for semantic segmentation under end-to-end architecture such as UNet, UNet++, DeepLabV3, and DeepLabv3+. Although, the already existing semantic segmentation methods for building detection work on the imagery of developed world, where the buildings are highly structured and there is a clearly distinguishable space present between the building instances, the same methods do not work as effectively on the developing world where there is often no clear differentiable spaces between instances of building thus reducing the number of detected instances. Hence as a noble approach, we have added building contours as new class along with building segmentation data, and detected the building contours and the inner building regions, hence giving the precise number of buildings existing in the input imagery especially in the convoluted areas where the boundary between the buildings are often hard to determine even for human eyes. Breaking down the building data into multi-class data increased the building detection precision and recall. This is useful in building detection where building instances are convoluted and are difficult for bare instance segmentation to detect all the instances.
基金supported by the Ministry for Innovation and Technology(Thematic Excellence Programme,TKP2020-NKA-04)and the NKFI K138079.
文摘Building extraction from remote sensing data is an important topic in urban studies and the deep learning methods have an increasing role due to their minimal requirements in training data to reach outstanding performance.We aimed to investigate the original U-Net architecture’s efficiency in building segmentation with different number of training images and the role of data augmentation based on multisource remote sensing data with varying spatial and spectral resolutions(WorldView-2[WV2],WorldView-3[WV3]images and an aerial orthophoto[ORTHO]).When the trainings and predictions were conducted on the same image,U-Net provided good results with very few training images(validation accuracies:94-97%;192 images).Combining the ORTHO’s and WV2’s training data for prediction on WV3 provided poor results with low F1-score(0.184).However,the inclusion of only 48 WV3 training images significantly improved the F1-score(0.693),thus,most buildings were correctly identified.Accordingly,using only independent reference data(other than the target image)is not enough to train an accurate model.In our case,the reference from WW2 and ORTHO images did not provide an acceptable basis to train a good model,but a minimal number of training images from the targeted WV3 improved the accuracy(F1-score:69%).
