Deep Learning Network Integrated Multi-spectral Data and Interferometric Imaging Radar Altimeter Data of Tiangong-2 for Land Use Classification
Multispectral data and radar data contain abundant spectral and texture features, which are widely used in land use classification. High precision registration and information mining are the two major difficulties faced by the classification of these two data. In this paper, visible and near infrared spectrum range of Wide-band Imaging Spectrometer (VNI) and Interferometric Imaging Radar Altimeterdata (InIRA) are used as raw data, and data registration and combination is carried out (the combination data is called for VNI_InIRA). Then the information of each pixel before and after combination is extracted and transformed into gray-scale image which is easy to be input by convolution neural network (CNN). Finally, the classification of land use is carried out by trained CNN models, and the accuracy is verified. The results show that the classification results of VNI_InIRA data are better than those of VNI data. The overall classification accuracy is 94.29% and 91.83% respectively. Therefore, the CNN which synthesizes the spectral characteristics of VNI data and the texture topographic features of InIRA data is a feasible method to obtain accurate land use classification information, and to provide a reference for the study of land use classification extraction in coastal areas.
KeywordsDeep learning Convolutional Neural Network Tiangong-2 Multi-spectral data Interferometric Imaging Radar Altimeter data Land use classification
Thanks to China Manned Space Engineering for providing space science and application data products of Tiangong-2. Thanks to the National R&D Infrastructure and Facility Development Program of China, “Fundamental Science Data Sharing Platform” (DKA2018-12-02-23) who funded the project.
- 1.Yang, H.X., Zhao, J.C., Gao, L., et al.: China’s space laboratory and manned space station. J. Nat. 69(4), 1–5 (2017)Google Scholar
- 5.Ji, Y.J., Yue, G.R., Zhang, W.F.: Use fusion of SAR and optical images for land cover classification. J. Southwest For. Univ. 36(3), 158–162 (2016)Google Scholar
- 6.Xu, W.: Classification of the intertidal saltmarsh using Sentinel-1 and Landsat-8 data. East China Normal University (2017)Google Scholar
- 7.Feng, L.Y.: Research on Construction Land Information Extraction from High Resolution Images with Deep Learning Technology. Zhejiang University (2017)Google Scholar
- 8.Qin, G.F.: Study of Multi-spectral Remote Sensing Imagery Classification and Urban Expansion Based on Machine Learning. Chongqing University (2012)Google Scholar
- 9.Zhang, H.L., Jiang, J.J., Wu, H.A., et al.: The BP neural network classification based on the fusion of SAR and TM images. Acta Geodaetica Cartogr. Sin. 35(3), 229–233 (2006)Google Scholar
- 10.Ye, C.: Study of Shallow Covering Geological Information based on the Polarized Wave Penetrability. Jilin University (2015)Google Scholar
- 11.He, K.M., Sun, J.: Convolutional neural networks at constrained time cost. In: Proceedings of the 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 5353–5360. IEEE, Boston (2015)Google Scholar
- 12.Wu, J.J., Li, Y.Z., Yu, L.J., et al.: Dynamic changes and driving factors of landscape connectivity for natural wetland in Yellow River Delta. Ecol. Environ. Sci. 27(1), 71–78 (2018)Google Scholar
- 13.Wang, X.H., Li, Y.Z., Meng, H., et al.: Distribution pattern of plant community in new-born coastal wetland in the yellow river delta. Scientia Geogr. Sin. 35(8), 1021–1026 (2015)Google Scholar