using numpy arrays for classification

[Nobody/Anonymous]

Hello,

I tried Spectral Python 0.6 to classify a tif file with the following code:

from osgeo import gdal
import numpy as np
import spectral as spy

landsat = gdal.Open('/tmp/L72192027_02720020914_B80_900913_cut.tif')
npLandsat = tif.ReadAsArray()

view(npLandsat)
(m, c) = spy.cluster(npLandsat, 5)

After the cluster i've got the following error:

---

IndexError                                Traceback (most recent call last)
/home/kaotika/<ipython-input-7-2e98755e86c6> in <module>()
---> 1 (m, c) = spy.cluster(npLandsat, 5)

/usr/local/lib/python2.6/dist-packages/spectral/algorithms/clustering.pyc in cluster(data, nClusters)
    481     '''
    482     opc = OnePassClusterer(nClusters)
-> 483     return opc.classifyImage(data)
    484
    485

/usr/local/lib/python2.6/dist-packages/spectral/algorithms/clustering.pyc in classifyImage(self, image)
    419         else:
    420             typecode = 'f'
-> 421         self.clusters = numpy.zeros((self.maxClusters, self.image.shape), typecode)
    422         self.nClusters = 0
    423         clusters = self.clusters

IndexError: tuple index out of range

Is it possible to classify (supervised and unsupervised) images imported as numpy array? Is it possible to convert numpy arrays to spy arrays?

Thanks for help

[Thomas Boggs]

Yes, it will work with numpy arrays.  spy is intended for use with multi-band images so the clustering algorithm expects the shape of its input to be (M, N, K), where M, N, and K are the numbers of rows, columns and bands, respectively.  My guess is that your array is only two dimensional (shape is (M, N)).  Try adding the third dimension to the shape:

npLandsat.shape = npLandsat.shape + (1,)

-thomas

[Nobody/Anonymous]

This was it. Now it works perfektly! Many Thanks!