Luo-IGARSS2011-2385.ppt

925 views

Published on

Published in: Technology, Education
0 Comments
1 Like
Statistics
Notes
  • Be the first to comment

No Downloads
Views
Total views
925
On SlideShare
0
From Embeds
0
Number of Embeds
11
Actions
Shares
0
Downloads
18
Comments
0
Likes
1
Embeds 0
No embeds

No notes for slide
  • The purposes of this work is to evaluate the unsupervised linear unmixing approaches for hs images for crop yield estimation To use the multidate data in order to improve the estimation results
  • The spectrum of a pixel of a hyperspectral image is considered as a linear mixture of the spectra of several endmembers In this paper we use VCA method to unmix hyperspectral images. It extracts the extremal points of the simplex formed by the date points as the spectra of endmembers.
  • For each field, the crop abundances are firstly estimated by VCA and NNLS from the hyperspectral images taken at different dates separately. The crop abundances of different dates are then combined by different ways, see the next page.
  • M1 is simply the crop abundances computed on the image of the date 18-May-2001 M2 is simply the crop abundances computed on the image of the date 29-May-2001 The crop yield estimation is evaluated by computing the correlation coefficients between the crop yield data and the (combined) crop abundances
  • Luo-IGARSS2011-2385.ppt

    1. 1. LINEAR UNMIXING OF MULTIDATE HYPERSPECTRAL IMAGERY FOR CROP YIELD ESTIMATION Bin Luo 1 , Chenghai Yang 2 and Jocelyn Chanussot 3 1 LIESMARS, Wuhan University, Wuhan, China 2 U.S. Department of Agriculture, Weslaco, Texas, USA 3 Grenoble Institute of Technology, Grenoble, France IGARSS 2011; 24 – 29 July, 2011; Vancouver, Canada
    2. 2. Mapping Yield Variation for Precision Agriculture <ul><li>Remote sensing imagery has been commonly used for estimating crop yield variation </li></ul><ul><li>Vegetation indices (e.g., NDVI) </li></ul><ul><li>With hyperspectral imagery, the number of VIs is large </li></ul><ul><li>Spectral unmixing can be used to derive abundance images </li></ul>
    3. 3. Spectral Mixing <ul><li>A pixel can be considered as a mixture of plants and soil. </li></ul><ul><li>Spectral unmixing can quantify crop canopy fraction within each pixel. </li></ul><ul><li>A crop fraction image is a more direct measure of plant abundance than NDVI </li></ul><ul><li>Plant abundance is indicative of crop yield. </li></ul>Plant Soil Mixture
    4. 4. Objectives and Procedures <ul><li>Evaluate unsupervised linear unmixing approaches on hyperspectral images for crop yield estimation </li></ul><ul><li>Use multi-date hyperspectral data for improving estimation results </li></ul>26-July-2011 Linear Unmixing of Multidate Hyperspectral Imagery for Crop Yield Estimation VCA ( Vertex Component Analysis
    5. 5. <ul><li>Linear mixture model of hyperspectral images </li></ul><ul><li>X = MS + n </li></ul><ul><li>M = unmixing matrix </li></ul><ul><li>S = abundance matrix </li></ul>Unmixing of Hyperspectral Images <ul><li>VCA ( Vertex Component Analysis ) to extract endmembers </li></ul><ul><li>Red cross: </li></ul><ul><li>hyperspecral data X </li></ul><ul><li>Blue circles: </li></ul><ul><li>endmembers M </li></ul><ul><li>Abundance S : </li></ul><ul><li>Random between 0 – 1 </li></ul>26-July-2011 Linear Unmixing of Multidate Hyperspectral Imagery for Crop Yield Estimation
    6. 6. Airborne Hyperspectral Images <ul><li>Hyperspectral system </li></ul><ul><ul><li>Spectral range: 467–932 nm </li></ul></ul><ul><ul><li>Swath width: 640 pixels </li></ul></ul><ul><ul><li>Bands: 128 </li></ul></ul><ul><ul><li>Radiometric: 12 bit (0–4095) </li></ul></ul><ul><ul><li>Pixel size: ~1 m </li></ul></ul><ul><li>Study site </li></ul><ul><ul><li>Two grain sorghum fields in south Texas </li></ul></ul><ul><ul><li>13.4 ha and 14.0 ha in size </li></ul></ul><ul><li>Image timing </li></ul><ul><ul><li>Shortly before and after crop reached maximum canopy cover </li></ul></ul><ul><ul><li>18-May-2001 and 29-May-2001 </li></ul></ul>Linear Unmixing of Multidate Hyperspectral Imagery for Crop Yield Estimation
    7. 7. Geometric Correction, Rectification & Calibration <ul><li>Geometric correction </li></ul><ul><ul><li>Reference line approach </li></ul></ul><ul><li>Rectification </li></ul><ul><ul><li>Georeference images to UTM </li></ul></ul><ul><ul><li>with GPS ground control points </li></ul></ul><ul><li>Radiometric calibration </li></ul><ul><ul><li>Three tarps with reflectance of 4, 32, and 48% were used to convert digital counts to reflectance </li></ul></ul><ul><li>102 bands were used for analysis </li></ul>Raw Corrected Linear Unmixing of Multidate Hyperspectral Imagery for Crop Yield Estimation
    8. 8. Grain Sorghum Yield Data Collection Ag Leader PF3000 Yield Monitor Linear Unmixing of Multidate Hyperspectral Imagery for Crop Yield Estimation
    9. 9. Yield Data Crop yield images of the two fields. 26-July-2011 Linear Unmixing of Multidate Hyperspectral Imagery for Crop Yield Estimation
    10. 10. Fusion of Multi-date Unmixing Results Flow chart of the fusion of the multi-date unmixing results 26-July-2011 Linear Unmixing of Multidate Hyperspectral Imagery for Crop Yield Estimation
    11. 11. <ul><li>M 18 (k) and M 29 (k ) as the abundances of crop extracted on the date 18 May 2001 and 29 May 2001 at the k th pixel </li></ul><ul><li>Evaluation – Correlation coefficients </li></ul>Fusion of Multi-date Unmixing Results where Y is the yield data 26-July-2011 Linear Unmixing of Multidate Hyperspectral Imagery for Crop Yield Estimation
    12. 12. Fusion of Multi-date Unmixing Results M 18 (k) of Field 1 M 29 (k) of Field 1 26-July-2011 Linear Unmixing of Multidate Hyperspectral Imagery for Crop Yield Estimation
    13. 13. Fusion of Multi-date Unmixing Results M 18 (k) of Field 2 M 29 (k) of Field 2 26-July-2011 Linear Unmixing of Multidate Hyperspectral Imagery for Crop Yield Estimation
    14. 14. Fusion of Multi-date Unmixing Results Correlation coefficients between the yield data and the (combined) crop abundances of Field 1 Correlation coefficients between the yield data and the (combined) crop abundances of Field 2 Recall that 26-July-2011 Linear Unmixing of Multidate Hyperspectral Imagery for Crop Yield Estimation M 1 M 2 M 3 M 4 C(M i , Y) 0.739 0.748 0.780 0.764 M 1 M 2 M 3 M 4 C(M i , Y) 0.648 0.721 0.735 0.701
    15. 15. Conclusions <ul><li>Crop abundances obtained by the unsupervised linear unmixing are strongly correlated to crop yield data. </li></ul><ul><li>The fusion of crop abundances obtained from images taken at different dates significantly improves the correlation with yield. </li></ul>26-July-2011 Linear Unmixing of Multidate Hyperspectral Imagery for Crop Yield Estimation

    ×