logo

SCIENTIA SINICA Informationis, Volume 51 , Issue 3 : 449(2021) https://doi.org/10.1360/SSI-2019-0229

Spectral dimensional correlation and sparse reconstruction model of hyperspectral images

More info
  • ReceivedOct 16, 2019
  • AcceptedFeb 18, 2020
  • PublishedFeb 24, 2021

Abstract


Funded by

国家自然科学基金项目(41671439,41971388)

辽宁省高等学校创新团队支持计划(LT2017013)


Acknowledgment

非常感谢匿名评审专家所提出的中肯而有建设性的修改意见.


References

[1] Liu D Z, Huang S Q, Wang Y T, et al. Hyperspectral remote sensing image processing and application. Beijing: Science Press. 2018. Google Scholar

[2] Zhang L P, Li J Y. Development and prospect of sparse representation-based hyperspectral image processing and analysis. J Remote Sens, 2016, 20: 1091--1101. Google Scholar

[3] Bioucas-Dias J M, Plaza A, Camps-Valls G. Hyperspectral Remote Sensing Data Analysis and Future Challenges. IEEE Geosci Remote Sens Mag, 2013, 1: 6-36 CrossRef Google Scholar

[4] Donoho D L. Compressed sensing. IEEE Trans Inform Theor, 2006, 52: 1289-1306 CrossRef Google Scholar

[5] Candes E J, Wakin M B. An Introduction To Compressive Sampling. IEEE Signal Process Mag, 2008, 25: 21-30 CrossRef ADS Google Scholar

[6] Jiao L C, Yang S Y, Liu F, et al. Development and prospect of compressive sensing. Acta Electron Sin, 2011, 39: 1651--1662. Google Scholar

[7] Shao W Z, Wei Z H. Advances and perspectives on compressed sensing theory. J Image Graph, 2012, 17: 1--12. Google Scholar

[8] Jihao Yin , Jianying Sun , Xiuping Jia . Sparse Analysis Based on Generalized Gaussian Model for Spectrum Recovery With Compressed Sensing Theory. IEEE J Sel Top Appl Earth Observations Remote Sens, 2015, 8: 2752-2759 CrossRef ADS Google Scholar

[9] Ghamisi P, Yokoya N, Li J. Advances in Hyperspectral Image and Signal Processing: A Comprehensive Overview of the State of the Art. IEEE Geosci Remote Sens Mag, 2017, 5: 37-78 CrossRef Google Scholar

[10] Qian Y, Ye M. Hyperspectral Imagery Restoration Using Nonlocal Spectral-Spatial Structured Sparse Representation With Noise Estimation. IEEE J Sel Top Appl Earth Observations Remote Sens, 2013, 6: 499-515 CrossRef ADS Google Scholar

[11] Xue J, Zhao Y, Liao W. Joint Spatial and Spectral Low-Rank Regularization for Hyperspectral Image Denoising. IEEE Trans Geosci Remote Sens, 2018, 56: 1940-1958 CrossRef ADS Google Scholar

[12] He Z, Li J, Liu L. Tensor Block-Sparsity Based Representation for Spectral-Spatial Hyperspectral Image Classification. Remote Sens, 2016, 8: 636 CrossRef ADS Google Scholar

[13] Yang J, Li Y, Chan J. Image Fusion for Spatial Enhancement of Hyperspectral Image via Pixel Group Based Non-Local Sparse Representation. Remote Sens, 2017, 9: 53-71 CrossRef ADS Google Scholar

[14] Wang Q, Ma L L, Tang L L, et al. Hyperspectral compressive sensing reconstruction based on spectral sparse model. J Infrared Millim Wave, 2016, 35: 723--730. Google Scholar

[15] Gelvez T, Rueda H, Arguello H. Joint sparse and low rank recovery algorithm for compressive hyperspectral imaging. Appl Opt, 2017, 56: 6785-6795 CrossRef ADS Google Scholar

[16] Xu P, Chen B, Xue L. A Prediction-Based Spatial-Spectral Adaptive Hyperspectral Compressive Sensing Algorithm. Sensors, 2018, 18: 3289 CrossRef Google Scholar

[17] Wang Y, Lin L, Zhao Q. Compressive Sensing of Hyperspectral Images via Joint Tensor Tucker Decomposition and Weighted Total Variation Regularization. IEEE Geosci Remote Sens Lett, 2017, 14: 2457-2461 CrossRef ADS Google Scholar

[18] Wang X H, Song H Y, Song C M, et al. Hyperspectral image compressed sensing model based on the collaborative sparsity of the intra-frame and inter-band. Sci Sin Inform, 2016, 46: 361--375. Google Scholar

[19] Zhang J, Zhao D, Zhao C. Image Compressive Sensing Recovery via Collaborative Sparsity. IEEE J Emerg Sel Top Circuits Syst, 2012, 2: 380-391 CrossRef ADS Google Scholar

[20] Li Y T. Research on hyperspectral image compression perception model based on multidimensional correlation. Dissertation for Master Degree. Dalian: Liaoning Normal University, 2018. Google Scholar

[21] Liwei Wang , Yan Zhang , Jufu Feng . On the Euclidean distance of images. IEEE Trans Pattern Anal Machine Intell, 2005, 27: 1334-1339 CrossRef Google Scholar

[22] Chen H D, Pu H Y, Wang B. Image Euclidean distance-based manifold dimensionality reduction algorithm for hyperspectral imagery. J Infrared Millimeter Waves, 2013, 32: 450-454 CrossRef Google Scholar

[23] Goldstein T, Osher S. The Split Bregman Method for L1-Regularized Problems. SIAM J Imag Sci, 2009, 2: 323-343 CrossRef Google Scholar

[24] Zhang J, Zhao D B. Split Bregman iteration based collaborative sparsity for image compressive sensing recovery. Intell Comput Appl, 2014, 4: 60--64. Google Scholar

[25] Lin Zhang , Lei Zhang , Xuanqin Mou . FSIM: A Feature Similarity Index for Image Quality Assessment. IEEE Trans Image Process, 2011, 20: 2378-2386 CrossRef ADS Google Scholar

[26] Figueiredo M A T, Nowak R D, Wright S J. Gradient Projection for Sparse Reconstruction: Application to Compressed Sensing and Other Inverse Problems. IEEE J Sel Top Signal Process, 2007, 1: 586-597 CrossRef ADS Google Scholar

[27] 候洵 . Hyperspectral Image Compressed Sensing Based on Linear Filter Between Bands. 光子学报, 2012, 41: 82-86 CrossRef Google Scholar

[28] Li C, Yin W, Zhang Y. TVAL3: TV minimization by augmented Lagrangian and alternating direction algorithms. 2013. http://www.caam.rice.edu/optimiza-tion/L1/TVAL3/. Google Scholar

  • Figure 1

    (Color online) Schematic diagrams of HSI spectral dimension plan

  • Figure 3

    (Color online) Part of the adjacent spectral curve segment in the 9th column spectral dimension

  • Figure 4

    The band image and the spectral dimension plan. (a) The 90th band image (Pavia); (b) the 90th column spectral dimension plan (Pavia); (c) the 95th band image (Chikusei); (d) the 95th column spectral dimension plan (Chikusei)

  • Figure 5

    (Color online) Comparison of texture features about the Pavia band image and the spectral dimension plan. (a) Contrast; (b) inverse different moment; (c) entropy; (d) homogeneity; (e) angular second moment

  • Figure 6

    (Color online) Comparison of texture features about the Chikusei band image and the spectral dimension plan. (a) Contrast; (b) inverse different moment; (c) entropy; (d) homogeneity; (e) angular second moment

  • Figure 9

    (Color online) Global nonlocal correlation diagram of HSI

  • Figure 10

    (Color online) Mining sparse representations in spectral dimension structure correlation

  • Figure 11

    Partial band images of HSI band group in experiment

  • Figure 12

    (Color online) Comparison of the reconstructed images PSNR in Dalian coastal zone of Hyperion between different algorithms at different sampling rates. Sampling rate of (a) 0.2, (b) 0.3, and (c) 0.4

  • Figure 13

    (Color online) Comparison of the reconstructed images PSNR in Cuprite of AVIRIS between different algorithms at different sampling rates. Sampling rate of (a) 0.2, (b) 0.3, and (c) 0.4

  • Figure 14

    (Color online) Comparison of the reconstructed images PSNR in Chikusei of Hyperspec between different algorithms at different sampling rates. Sampling rate of (a) 0.2, (b) 0.3, and (c) 0.4

  • Figure 15

    (Color online) Comparison of the reconstructed images PSNR in Pavia of Rosis-3 between different algorithms at different sampling rates. Sampling rate of (a) 0.2, (b) 0.3, and (c) 0.4

  • Table 1   Spectral angle statistics of spectral curve and its adjacent spectral curve in spectral dimension plan
    Spectral dimension plan number Reference spectral vector number ${k}$ Spectral angle between reference spectral vector $t_{k}$ and adjacent spectral vector $t_{k~\pm~i}$ (${i}=1,2,3,4$)
    $t_{k-4}$ $t_{k-3}$ $t_{k-2}$ $t_{k-1}$ $~~t_{k}$ $t_{k+1}$ $t_{k+2}$ $t_{k+3}$ $t_{k+4}$
    9 50 0.73 0.55 0.94 0.37 0 0.44 0.68 0.67 0.59
    100 1.61 0.75 1.48 0.87 0 0.31 1.81 1.01 1.56
    150 0.4 1.49 1.8 0.44 0 0.45 0.65 0.92 0.71
    200 0.57 1.08 0.55 0.54 0 1.14 0.58 0.53 0.78
    61 50 0.9 0.76 0.77 0.6 0 0.77 1.12 1.18 1.44
    100 0.94 1.32 1.07 0.89 0 0.85 0.73 0.96 1.28
    150 1.11 0.78 0.63 0.71 0 0.58 0.36 0.43 0.62
    200 1.96 1.49 1.12 1.34 0 0.68 1.73 1.26 1.1
    113 50 1.53 2.33 1.81 0.99 0 0.66 1 1.14 1.52
    100 2.53 0.62 1 0.67 0 0.76 0.54 0.57 1.61
    150 0.86 1.56 0.66 0.94 0 0.63 1.05 1.41 1.39
    200 2.24 1.48 1.02 1.44 0 2.81 2.6 3.81 3.71
    165 50 1.45 1.58 0.65 1.32 0 1.45 0.95 2.57 1.78
    100 1.37 1.24 0.98 1.23 0 2.23 0.89 0.7 1.12
    150 1.52 0.89 1.27 0.8 0 0.89 1.6 0.95 1.19
    200 0.97 0.66 0.9 0.81 0 1.82 1.47 1.42 0.78
  • Table 2   Statistics on reference blocks and similar spectral curve blocks in the HSI spectral dimension of the coastal zone in Dalian
    Reference block location
    (72, 80) (120, 87) (37, 48) (106, 34) (142, 73) (203, 76)

    Similar block

    information

    in the overall search area

    1 Location (72, 75) (120, 77) (37, 37) (106, 40) (142, 81) (203, 80)
    Similarity 0.0231 0.0327 0.0578 0.0222 0.0166 0.0235
    Distance from reference block 5 10 11 6 8 4
    2 Location (72, 74) (120, 65) (37, 57) (106, 39) (142, 80) (203, 81)
    Similarity 0.0259 0.0331 0.0579 0.0236 0.0193 0.0238
    Distance from reference block 6 22 9 5 7 5
    3 Location (72, 76) (120, 82) (37, 79) (106, 41) (142, 82) (203, 86)
    Similarity 0.0277 0.0347 0.0591 0.0244 0.02 0.0252
    Distance from reference block 4 5 31 7 9 10
    4 Location (72, 73) (120, 81) (37, 54) (106, 38) (142, 85) (203, 82)
    Similarity 0.0289 0.0349 0.0594 0.0255 0.021 0.0262
    Distance from reference block 7 6 6 4 9 6
    5 Location (72, 72) (120, 80) (37, 58) (106, 42) (142, 84) (203, 84)
    Similarity 0.031 0.035 0.0597 0.0276 0.0212 0.0264
    Distance from reference block 8 7 10 8 11 8
  • Table 3   Comparison of the reconstructed images FSIM and SAM in Dalian coastal zone of Hyperion at different sampling rates
    Algorithm Sampling rate
    0.2 0.3 0.4
    FSIM SAM ($^{\circ}$) FSIM SAM ($^{\circ}$) FSIM SAM ($^{\circ}$)
    GPSR 0.58 24.344 0.62 20.4333 0.77 11.786
    SLF_GPSR 0.66 13.6745 0.78 7.7091 0.87 6.2841
    TV 0.77 6.0196 0.84 4.9002 0.87 4.7052
    HiCoSM 0.86 4.5998 0.9 4.3294 0.93 3.5475
    S-SCoSM 0.97 4.2413 0.98 3.5904 0.98 2.9543
  • Table 4   Comparison of the reconstructed images FSIM and SAM in Cuprite of AVIRIS at different sampling rates
    Algorithm Sampling rate
    0.2 0.3 0.4
    FSIM SAM ($^{\circ}$) FSIM SAM ($^{\circ}$) FSIM SAM ($^{\circ}$)
    GPSR 0.56 3.5810 0.80 0.8509 0.83 0.6925
    SLF_GPSR 0.79 1.5163 0.84 1.0139 0.87 0.9597
    TV 0.82 0.9421 0.85 0.5742 0.88 0.5555
    HiCoSM 0.88 0.9616 0.92 0.9105 0.95 0.7801
    S-SCoSM 0.99 0.7379 0.99 0.6720 0.99 0.6083
  • Table 5   Comparison of the reconstructed images FSIM and SAM in Chikusei of Hyperspec at different sampling rates
    Algorithm Sampling rate
    0.2 0.3 0.4
    FSIM SAM ($^{\circ}$) FSIM SAM ($^{\circ}$) FSIM SAM ($^{\circ}$)
    GPSR 0.61 2.3092 0.68 1.0652 0.83 0.5601
    SLF_GPSR 0.71 6.0992 0.84 1.3545 0.88 1.0283
    TV 0.79 0.7649 0.84 0.3667 0.87 0.3451
    HiCoSM 0.90 1.2493 0.95 0.8275 0.96 0.9042
    S-SCoSM 0.99 0.6703 0.99 0.5683 0.99 0.4791
  • Table 6   Comparison of the reconstructed images FSIM and SAM in Pavia of Rosis-3 at different sampling rates
    Algorithm Sampling rate
    0.2 0.3 0.4
    FSIM SAM ($^{\circ}$) FSIM SAM ($^{\circ}$) FSIM SAM ($^{\circ}$)
    GPSR 0.65 10.5909 0.69 4.5353 0.79 2.8592
    SLF_GPSR 0.71 8.8390 0.80 3.3184 0.85 2.1311
    TV 0.80 1.7677 0.84 1.1624 0.87 1.0814
    HiCoSM 0.90 1.3796 0.94 1.2262 0.96 1.1571
    S-SCoSM 0.99 1.2487 0.99 1.0745 0.99 0.9352
qqqq

Contact and support