kgc2010

KGC is an unsupervised classification algorithm authored by Ashlin Richardson and developed in partnership with David Goodenough, Hao Chen and the Canadian Forest Service's Advanced Forest Technologies (AFT) group team. This implementation was developed between April 1 -- August 19th, 2010 for presentation at IGARSS 2010 while the author was funded by the U.Vic Math & Stats department, and is occasionally still being improved. Graphics functions adapted from Nigel Stewart [2].

N.B. only Linux is supported

Firescar Delineation: 2002 Keg River Wildfire, Alberta, Canada (Example 1)

Thanks to JAXA for L-band fully-polarimetric SAR data. Shown: cluster selection from 3-d cluster plot window. The 3-d coordinates are the same as those used in the RGB encoding in the Image window. The color encoding and the 3-d coordinates for plotting are tied together in this implementation but can be switched on the fly to explore higher-dimensional data. A "very large" number of bands is supported.

To run Example 1:

  ./setup.sh
  make
  ./run.sh 

Land cover with Sentinel-2 (Example 2)

Shown: result with water class selected.

To run:

  ./run2.sh 

Instructions

3d cluster plot

• Zoom: hold down both mouse buttons and drag
• Pan: hold down right mouse button and drag
• Rotate: hold down left mouse button and drag

Click on cluster centres (blue) to visualize cluster in image domain (class window). This produces a binary classification

• after producing a binary classication, can press c (lower case) to switch back to multi-class view (in which pixels are coloured according to the mode to which they are assigned). In this representation (same as before a particular cluster is selected) the mode is the value of the "hilltop" determined

Image plot

Band selection: [ add photo for switching bands]

• type: r1 (then press return) to set band 1 as red
• type g2 (then press return) to set band 2 as green
• type b3 (then press return) to set band 3 as blue Similarly, type r3 g2 b1 for (red, green, blue) = (3, 2, 1) instead..

biblio

[1] Unsupervised Nonparametric Classification of Polarimetric SAR Data Using The K-nearest Neighbor Graph, A. Richardson, D. G. Goodenough, H. Chen, G. Hobart, B. Moa, W. Myrvold, proc. IEEE IGARSS, Honolulu, Hawaii, July 2010.

[2] gltzpr by Nigel Stewart http://www.nigels.com/glt/gltzpr/zpr.h

[3] Hierarchical Unsupervised Nonparametric Classification of Polarimetric SAR Time Series Data, A. Richardson, D. Goodenough, H. Chen, proc. IEEE IGARSS, Québec City, Canada, July 2014

[4] Mapping Fire Scars Using Radarsat-2 Polarimetric SAR Data, D. Goodenough, H. Chen, A. Richardson, S. Cloude, W. Hong, Y. Li, Can. J. Remote Sensing, Vol. 37, No. 5, pp. 1-10, 2011

Various easy improvements that could be made

Changing ball sizes, adding some of the earlier array of density estimation of formulas, cluster merging / hierarchical operation / dendrogram.