Research on small building extraction method in small dataset

doi:10.13474/j.cnki.11-2246.2019.0317

Abstract

Abstract: The remote sensing satellite images are generally large, while the area containing small buildings is relatively small. If sliding cropping is used to augment data, most of the cropped patches have no targets, and a large dataset containing a large number of small buildings is time-consuming and labor intensive to construct. It is very difficult for conventional methods to extract small buildings on high-resolution satellite imagery and of great theoretical significance and application value to study the small building extraction method suitable for small-scale datasets. In this paper, a new lightweight and fully connected network ZF-FCN is proposed. ZF-FCN uses a small receptive field to obtain more local information, uses less max pooling operations to avoid violent down-sampling, and uses Lovász-Softmax loss to solve sample imbalance problem and better optimizes the IoU metric. A dataset consisted of a small number of images containing mainly small buildings is established in this paper, and the experiment is carried out on the augmented dataset with different cropping sizes, which finally proves that ZF-FCN outperforms both FCN and U-Net.

Key words: building extraction, fully connected network, receptive field, semantic segmentation, up-sampling

CLC Number:

P237

YANG Xubo, TIAN Jinwen. Research on small building extraction method in small dataset[J]. 测绘通报, 2019, 0(10): 51-55.

References 15

[1]	安文, 杨俊峰, 赵羲, 等. 高分辨率遥感影像的建筑物自动提取[J]. 测绘科学, 2014, 39(11):80-84.
[2]	侯蕾, 尹东, 尤晓建. 一种遥感图像中建筑物的自动提取方法[J]. 计算机仿真, 2006, 23(4):184-187.
[3]	SHELHAMER E, LONG J, DARRELL T. Fully convolutional networks for semantic segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Boston:IEEE, 2015:3431-3440.
[4]	RONNEBERGER O, FISCHER P, BROX T. U-net:convolutional networks for biomedical image segmentation[C]//International Conference on Medical Image Computing and Computer-assisted Intervention.[S.l.]:Springer, 2015:234-241.
[5]	刘文涛, 李世华, 覃驭楚. 基于全卷积神经网络的建筑物屋顶自动提取[J]. 地球信息科学学报, 2018, 20(11):1562-1570.
[6]	伍广明, 陈奇, SHIBASAKI R, 等. 基于U型卷积神经网络的航空影像建筑物检测[J]. 测绘学报, 47(6):864-872.
[7]	ZEILER M D, FERGUS R. Visualizing and understanding convolutional networks[C]//European Conference on Computer Vision. Zürich:Springer, 2014:818-833.
[8]	SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[C]//Proceedings of International Conference on ICLR.[S.l.]:arXiv, 2014.
[9]	IOFFE S, SZEGEDY C. Batch normalization:accelerating deep network training by reducing internal covariate shift[C]//Proceedings of the 32nd International Conference on Machine Learning.[S.l.]:arXiv, 2015.
[10]	ODENA A, DUMOULIN V, OLAH C. Deconvolution and checkerboard artifacts[J]. Distill, 2016, 1(10):e3.
[11]	RAKHLIN A, DAVYDOW A, NIKOLENKO S. Land cover classification from satellite imagery with U-Net and Lovász-Softmax loss[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). Salt Lake City:IEEE, 2018:257-2574.
[12]	BERMAN M, TRIKI R A, BLASCHKO M B. The Lovász-Softmax loss:a tractable surrogate for the optimization of the intersection-over-union measure in neural networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Salt Lake City:IEEE, 2018:4413-4421.
[13]	CONGALTON R G. A comparison of sampling schemes used in generating error matrices for assessing the accuracy of maps generated from remotely sensed data[J]. Photogrammetric Engineering and Remote Sensing, 1988,54(5):593-600.
[14]	QIAN N. On the momentum term in gradient descent learning algorithms[J]. Neural Networks, 1999, 12(1):145-151.
[15]	GLOROT X, BENGIO Y. Understanding the difficulty of training deep feedforward neural networks[C]//Proceedings of the 13th International Conference on Artificial Intelligence And Statistics. Sardinia:[s.n.], 2010:249-256.