Anterior|Siguiente

CNN with coefficient of variation-based dimensionality reduction for hyperspectral remote sensing images classification

  • role： First author第一作者
  • Affiliation：

    Technology and Engineering Center for Space Utilization, Chinese Academy of Sciences, Beijing 100094, China

    Key Laboratory of Space Utilization, Chinese Academy of Sciences, Beijing 100094, China

    University of Chinese Academy of Sciences, Beijing 100049, China

  • Email：zhangkang1128@126.com
  • Introduction：张康(1992—　)，女，硕士研究生，研究方向为遥感图像处理、机器学习等。E-mail：zhangkang1128@126.com
Kang ZHANG123,  
  • role： Corresponding author通信作者
  • Affiliation：

    Technology and Engineering Center for Space Utilization, Chinese Academy of Sciences, Beijing 100094, China

    Key Laboratory of Space Utilization, Chinese Academy of Sciences, Beijing 100094, China

  • Email：heibq@csu.ac.cn
  • Introduction：黑保琴(1975—　)，女，高级工程师，研究方向为信号与信息处理。E-mail：heibq@csu.ac.cn
Baoqin HEI12*,  
  • Affiliation：

    Technology and Engineering Center for Space Utilization, Chinese Academy of Sciences, Beijing 100094, China

    Key Laboratory of Space Utilization, Chinese Academy of Sciences, Beijing 100094, China

Zhuang ZHOU12,  
  • Affiliation：

    Technology and Engineering Center for Space Utilization, Chinese Academy of Sciences, Beijing 100094, China

    Key Laboratory of Space Utilization, Chinese Academy of Sciences, Beijing 100094, China

Shengyang LI12

resumen

Hyperspectral remote Sensing Images (HSIs), which contain rich spectral and spatial information, are important in the precise classification of earth objects. However, HSIs are usually highly dimensional and non-linear, and they contain large amounts of data. These characteristics increase the difficulty of data processing and bring about the Hughes phenomenon. Conventional methods such as Neural Networks (NN) and support vector machine have solved these problems by reducing the dimensions of HSIs with PCA, ICA, or MNF. Although these methods are effective in the classification of HSIs, they may cause information loss in the original data. Therefore, improving the accuracy of HSI classification with inadequate data is difficult. Recently, deep learning method, especially Convolutional Neural Network (CNN), has achieved remarkable performance in many fields. Therefore, the application of CNNs to HSI classification shows immense potential. To avoid the Hughes phenomenon and improve the accuracy of his classification, this study proposes a Coefficient of Variation–Convolution Neural Network (CV-CNN) method for the classification of HSIs. After the calculation of the Coefficient of Variation of the IntrA-class (CVIA) and the Coefficient of Variation of the IntEr-class (CVIE) of each band, the bands with low (CVIE)2/CVIA values are excluded. Then, a target pixel with the spectral information of its eight neighbors is organized as multi-layer spectral–spatial information. The spectral–spatial information of the target pixel should then be converted into matrix form. The two-dimensional image suitable for the input of CNN was subsequently obtained. Furthermore, a seven-layer CNN model was constructed with two convolution layers, two max-pooling layers, two full-connection layers, and one softmax layer. Using the seven-layer CNN can effectively improve the accuracy of HSI classification. Experiments were conducted on the Indian Pines dataset and Pavia University dataset to evaluate the performance of the presented CV-CNN method. The results are as follows: (1) Compared with the method that only considers spectral information for HSI classification, the use of spectral–spatial information can actually improve the accuracy of HSI classification. (2) Compared with other CNN methods, the seven-layer CNN model with no band removed can increase the overall accuracy by 2.61% for the Indian Pines dataset and 0.04% for the Pavia University dataset. (3) Based on the same seven-layer model, the experiments show that the classification of HSI with poor bands excluded shows increased accuracy from 97.9% to 98.69% for the Indian Pines dataset and from 99.6% to 99.66% for the Pavia University dataset. This outcome verifies the validity of excluding poor bands on the basis of the CV method. (4) Based on the same seven-layer model, the experiments show that the CV method, compared with the PCA and MNF methods, can increase the overall accuracy by 1.38% and 0.44%, respectively, for the Indian Pines dataset and by 3.26% and 1.83%, respectively, for the Pavia University dataset. The CNN model developed in this study and the method of removing poor bands with the CV technique can improve the accuracy of HSI classification.

palabra clave

Convolutional Neural Networks(CNN);deep learning;coefficient of variation;hyperspectral remote sensing images;classification

References

  1. 1.
    Bioucas-Dias J M, Plaza A, Camps-Valls G, Scheunders P, Nasrabadi N and Chanussot J. 2013. Hyperspectral remote sensing data analysis and future challenges. IEEE Geoscience and Remote Sensing Magazine, 1(2): 6–36 10.1109/MGRS.2013.2244672
  2. 2.
    Chen Y S, Jiang H L, Li C Y, Jia X P and Ghamisi P. 2016. Deep feature extraction and classification of hyperspectral images based on convolutional neural networks. IEEE Transactions on Geoscience and Remote Sensing, 54(10): 6232–6251 10.1109/TGRS.2016.2584107
  3. 3.
    Chen Y S, Lin Z H, Zhao X, Wang G and Gu Y F. 2014. Deep learning-based classification of hyperspectral data. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 7(6): 2094–2107 10.1109/JSTARS.2014.2329330
  4. 4.
    Feng J F and Brown D. 1998. Spike output jitter, mean firing time and coefficient of variation. Journal of Physics A: Mathematical and General, 31(4): 1239–1252 10.1088/0305-4470/31/4/013
  5. 5.
    Gao H M, Li C M, Zhou H, Zhang Z, Chen L H and He Z Y. 2016. Dimension reduction and classification of hyperspectral remote sensing images based on sensitivity analysis of artificial neural network. Journal of Electronics and Information Technology, 38(11): 2715–2723 (
  6. 6.
    Gevaert C M, Tang J, García-Haro F J and Kooistra L. 2014. Combining hyperspectral UAV and multispectral formosat-2 imagery for precision agriculture applications//Proceedings of the 6th Workshop on Hyperspectral Image and Signal Processing: Evolution in Remote Sensing. Lausanne, Switzerland: WHISPERS
  7. 7.
    He K M and Sun J. 2015. Convolutional neural networks at constrained time cost//Proceedings of the 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Boston, MA: IEEE: 5353–5360 [DOI: 10.1109/CVPR.2015.7299173]
  8. 8.
    Hu W, Huang Y Y, Wei L, Zhang F and Li H C. 2015. Deep convolutional neural networks for hyperspectral image classification. Journal of Sensors, 2015: 258619 10.1155/2015/258619
  9. 9.
    Krizhevsky A, Sutskever I and Hinton G E. 2012. ImageNet classification with deep convolutional neural networks//Proceedings of the 25th International Conference on Neural Information Processing Systems. Lake Tahoe, Nevada: Curran Associates Inc.: 1097–1105.
  10. 10.
    Kuo B C, Ho H H, Li C H, Hung C C and Taur J S. 2014. A kernel-based feature selection method for SVM with RBF kernel for hyperspectral image classification. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 7(1): 317–326 10.1109/JSTARS.2013.2262926
  11. 11.
    Lecun Y, Bottou L, Bengio Y and Haffner P. 1998. Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86(11): 2278–2324 10.1109/5.726791
  12. 12.
    Leng J B, Li T, Bai G, Dong Q K and Dong H. 2016. Cube-CNN-SVM: a novel hyperspectral image classification method//Proceedings of the 28th International Conference on Tools with Artificial Intelligence (ICTAI). San Jose, CA: IEEE: 1027–1034 [DOI: 10.1109/ICTAI.2016.0158]
  13. 13.
    Li H, Xiao G R, Xia T, Tang Y Y and Li L Q. 2014. Hyperspectral image classification using functional data analysis. IEEE Transactions on Cybernetics, 44(9): 1544–1555 10.1109/TCYB.2013.2289331
  14. 14.
    Liang L, Yang M H and Li Y F. 2010. Hyperspectral remote sensing image classification based on ICA and SVM algorithm. Spectroscopy and Spectral Analysis, 30(10): 2724–2728 (
  15. 15.
    Lu Y C, Hu C M, Sun S J, Zhang M W, Zhou Y, Shi J and Wen Y S. 2016. Overview of optical remote sensing of marine oil spills and hydrocarbon seepage. Journal of Remote Sensing, 20(5): 1259–1269 (
  16. 16.
    Mei S H, Ji J Y, Bi Q Q, Hou J H, Du Q and Li W. 2016. Integrating spectral and spatial information into deep convolutional Neural Networks for hyperspectral classification//Proceedings of the 2016 IEEE International Geoscience and Remote Sensing Symposium. Beijing, China: IEEE: 5067–5070 [DOI: 10.1109/IGARSS.2016.7730321]
  17. 17.
    Roberts D A, Quattrochi D A, Hulley G C, Hook S J and Green R O. 2012. Synergies between VSWIR and TIR data for the urban environment: an evaluation of the potential for the Hyperspectral Infrared Imager (HyspIRI) Decadal Survey mission. Remote Sensing of Environment, 117: 83–101 10.1016/j.rse.2011.07.021
  18. 18.
    Tan X, Yu X C, Qin J C and Wei X P. 2014. Multiple kernel SVM classification for hyperspectral images. Chinese Journal of Scientific Instrument, 35(2): 405–411 (
  19. 19.
    Tian Y P, Tao C, Zou Z R, Yang Z X and He X F. 2015. Semi-supervised graph-based hyperspectral image classification with active learning. Acta Geodaetica et Cartographica Sinica, 44(8): 919–926 (
  20. 20.
    Tiwari S K, Saha S K and Kumar S. 2015. Prediction modeling and mapping of soil carbon content using artificial neural network, hyperspectral satellite data and field spectroscopy. Advances in Remote Sensing, 4(1): 63–72 10.4236/ars.2015.41006
  21. 21.
    Zhang C Y, Cheng H F, Chen Z H and Zheng W W. 2008. The development of hyperspectral remote sensing and its threatening to military equipments. Electro-Optic Technology Application, 23(1): 10–12 (
  22. 22.
    Zhang C Y, Qin Q M, Chen L, Wang N and Zhao S S. 2015. Research and development of mineral identification utilizing hyperspectral remote sensing. Optics and Precision Engineering, 23(8): 2407–2418 (

Leer el texto completo

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