предыдущий|следующий

House building extraction from high-resolution remote sensing images based on IEU-Net

  • role： First author第一作者
  • Affiliation：

    Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China

    University of Chinese Academy of Sciences, Beijing 100049, China

  • Email：wangzhenqing19@mails.ucas.ac.cn
  • Introduction：王振庆， 1997年生， 男， 博士研究生， 研究方向为深度学习语义分割与目标检测。E-mail： wangzhenqing19@mails.ucas.ac.cn
WANG Zhenqing12,  
  • role： Corresponding author通信作者
  • Affiliation：

    Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China

  • Email：zhouyi@radi.ac.cn
  • Introduction：周艺， 1964年生， 女， 研究员， 研究方向为人居环境遥感应用。E-mail： zhouyi@radi.ac.cn
ZHOU Yi1*,  
  • Affiliation：

    Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China

WANG Shixin1,  
  • Affiliation：

    Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China

WANG Futao1,  
  • Affiliation：

    Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China

    University of Chinese Academy of Sciences, Beijing 100049, China

XU Zhiyu12

реферат

House buildings, as the main place for human activities, can be rapidly and accurately extracted from high-resolution remote sensing images, which is of great significance for promoting remote sensing information in disaster prevention and mitigation and town management. On the basis of deep learning, this paper proposes a pixel-level accurate extraction method for house buildings by using high-resolution remote sensing images.First, in consideration of the lack of pixel features at the edge of a sample image, an IEU-Net model is proposed on the basis of the U-Net model. A new ignore-edges categorical cross entropy function, IELoss, is designed as a loss function. Dropout and BN layers are added to help improve the speed and robustness of model training while fitting. Second, to solve the problem of limited feature richness of the model, the Morphological Building Index (MBI) is introduced into the classification process of the model, together with the remote sensing image RGB band. Lastly, when ignoring the edge prediction method, the best building extraction results can be obtained in the model prediction that corresponds to the IEU-Net model. The remote sensing image data used in this study are 0.8 m true color and infrared band data obtained by multispectral and panchromatic fusion from the GF-2 satellite of the Yushu Tibetan Autonomous Prefecture of Qinghai Province. A label image is obtained by visually interpreting the images by using ArcGIS software and performing a feature-to-raster transformation on the obtained vector file. Then, a ground truth image of the location of houses and buildings in the remote sensing image is recorded. The MBI and RGB bands are fused, and the corresponding label images are randomly cropped to obtain 500 pieces of 256×256 training data and 100 pieces of 256×256 verification data. Regarding the IEU-Net model r value of 0.5 and the training data as the model input and using the Adam optimization algorithm for back propagation iteration for 100 times help achieve the optimal parameter results.The Overall Accuracy (OA) is 91.86%, and the kappa value is 0.802. To verify the effectiveness of IELoss in the IEU-Net model and the ignore-edges prediction method relative to CELoss and the ordinary prediction method in solving the problem of insufficient edge pixel features, this study continues the comparison by using the values of r of 1.0, 0.9, 0.8, 0.7, and 0.6 (when r is 1.0, IELoss is equivalent to CELoss, and the ignore-edges prediction method is equivalent to the ordinary prediction method). The results show that the OA of the model using IELoss and the ignore-edges prediction method is 5.03% higher than that of the model using CELoss and the ordinary prediction method, and the kappa value is 0.165 higher. To verify the effectiveness of adding MBI to the training data of the model, a comparative experiment is performed by using RGB three-band data as the training set. The prediction result obtained using the MBI-added data set is 1.55% higher than the OA of the RGB band data set. The kappa value is increased by 0.009.The experimental results show that the IEU-Net model effectively solves the problem of insufficient edge pixel features and achieves house and building extraction accuracy with high OA and kappa values. The addition of MBI data can overcome the influence of roads and house building shadows to a certain extent and accurately extract the edge information of house buildings. Therefore, if significant vegetation exists near the houses and buildings in our area of interest, we can consider adding the normalized difference vegetation index to the training set to reduce the impact of the vegetation.

ключеви́че слова́

deep learning;IEU-Net;high-resolution remote sensing image;House building;MBI

References

  1. 1.
    Audebert N, Le Saux B and Lefèvre S. 2017. Semantic segmentation of earth observation data using multimodal and multi-scale deep networks//13th Asian Conference on Computer Vision. Taipei, China: Springer: 180-196
  2. 2.
    Chen K Q, Fu K, Yan M L, Gao X, Sun X and Wei X. 2018. Semantic segmentation of aerial images with shuffling convolutional neural networks. IEEE Geoscience and Remote Sensing Letters, 15(2): 173-177
  3. 3.
    Chen L C, Zhu Y K, Papandreou G, Schroff F and Adam H. 2018. Encoder-decoder with atrous separable convolution for semantic image segmentation//15th European Conference on Computer Vision. Munich: Springer: 833-851
  4. 4.
    Fan R S, Chen Y, Xu Q H and Wang J X. 2019. A high-resolution remote sensing image building extraction method based on deep learning. Acta Geodaetica et Cartographica Sinica, 48(1): 34-41
  5. 5.
    Galvanin E and Dal Poz A P P. 2012. Extraction of building roof contours from LiDAR data using a Markov-random-field-based approach. IEEE Transactions on Geoscience and Remote Sensing, 50(3): 981-987
  6. 6.
    Huang X and Zhang L P. 2011. A multidirectional and multiscale morphological index for automatic building extraction from multispectral geoeye-1 imagery. Photogrammetric Engineering and Remote Sensing, 77(7): 721-732
  7. 7.
    Ioffe S and Szegedy C. 2015. Batch normalization: accelerating deep network training by reducing internal covariate shift//Proceedings of the 32nd International Conference on International Conference on Machine Learning. Lille: JMLR.org: 448-456
  8. 8.
    Jin X Y and Davis C H. 2005. Automated building extraction from high-resolution satellite imagery in urban areas using structural, contextual, and spectral information. EURASIP Journal on Advances in Signal Processing, 2005(14): 2196-2206
  9. 9.
    Kabolizade M, Ebadi H and Ahmadi S. 2010. An improved snake model for automatic extraction of buildings from urban aerial images and LiDAR data. Computers, Environment and Urban Systems, 34(5): 435-441
  10. 10.
    Kang W C, Xiang Y M, Wang F and You H J. 2019. EU-Net: an efficient fully convolutional network for building extraction from optical remote sensing images. Remote Sensing, 11(23): 2813
  11. 11.
    Mnih V. 2013. Machine Learning for Aerial Image Labeling. Toronto: University of Toronto
  12. 12.
    Ronneberger O, Fischer P and Brox T. 2015. U-Net: convolutional networks for biomedical image segmentation//18th International Conference on Medical Image Computing and Computer-Assisted Intervention. Munich: Springer: 234-241
  13. 13.
    Saito S, Yamashita T and Aoki Y. 2016. Multiple object extraction from aerial imagery with convolutional neural networks//Electronic Imaging, Intelligent Robots and Computer Vision XXXIII: Algorithms and Techniques. [s.l.]: Society for Imaging Science and Technology: 1-9
  14. 14.
    Shelhamer E, Long J and Darrell T. 2017. Fully convolutional networks for semantic segmentation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 39(4): 640-651
  15. 15.
    Song J E, Wang X and Li P J. 2012 Urban building damage detection from VHR imagery by including temporal and spatial texture features. Journal of Remote Sensing, 16(6): 1233-1245
  16. 16.
    Srivastava N, Hinton G, Krizhevsky A, Sutskever I and Salakhutdinov R. 2014. Dropout: a simple way to prevent neural networks from overfitting. The Journal of Machine Learning Research, 15(1): 1929-1958
  17. 17.
    Sun X, Fu K, Long H, Hu Y F, Cai L and Wang H Q. 2009. Contextual models for automatic building extraction in high resolution remote sensing image using object-based boosting method//IEEE International Geoscience and Remote Sensing Symposium. Boston: IEEE: II-437-II-440
  18. 18.
    Tao C, Tan Y H, Cai H J, Du B and Tian J W. 2010. Object-oriented method of hierarchical urban building extraction from high-resolution remote-sensing imagery. Acta Geodaetica et Cartographica Sinica, 39(1): 39-45
  19. 19.
    Tian J J and Reinartz P. 2013. Fusion of multi-spectral bands and DSM from Worldview-2 Stereo imagery for building extraction//Joint Urban Remote Sensing Event. Sao Paulo: IEEE: 135-138
  20. 20.
    Turker M and Koc-San D. 2015. Building extraction from high-resolution optical spaceborne images using the integration of support vector machine (SVM) classification, hough transformation and perceptual grouping. International Journal of Applied Earth Observation and Geoinformation, 34: 58-69
  21. 21.
    You Y F, Wang S Y, Wang B, Ma Y X, Shen M, Liu W H and Xiao L. 2019. Study on hierarchical building extraction from high resolution remote sensing imagery. Journal of Remote Sensing, 23(1): 125-136

Читать полностью

The above content is generated by Large Model Translation. The translated content is for reference only. We do not assume any commercial or legal responsibilty for any consequences arising from the use of our website