def save_ice_map(inp_filename, raw_filename, classifier_filename, threads,
source, quicklook, force):
""" Load texture features, apply classifier and save ice map """
# get filenames
out_filename = inp_filename.replace('_texture_features.npz',
'_classified_%s.tif' % source)
if os.path.exists(out_filename) and not force:
print('Processed file %s already exists.' % out_filename)
return out_filename
# import classifier
clf = pickle.load(open(classifier_filename, "rb"))
clf.n_jobs = threads
# get texture features
npz = np.load(inp_filename)
features = np.vstack([
npz['textureFeatures'].item()['HH'],
npz['textureFeatures'].item()['HV'],
])
imgSize = features.shape[1:]
features = features.reshape((26, np.prod(imgSize))).T
gpi = np.isfinite(features.sum(axis=1))
result = clf.predict(features[gpi, :])
classImage = np.ones(np.prod(imgSize)) * 255
classImage[gpi] = result
classImage = classImage.reshape(imgSize)
img_shape = classImage.shape
# open original file to get geometry
raw_nansat = Nansat(raw_filename)
# crop and resize original Nansat to match the ice map
raw_shape = raw_nansat.shape()
crop = [rshape % ishape for (rshape, ishape) in zip(raw_shape, img_shape)]
raw_nansat.crop(0, 0, raw_shape[1] - crop[1], raw_shape[0] - crop[0])
raw_nansat.resize(height=img_shape[0])
raw_nansat.reproject_gcps()
# create new Nansat object and add ice map
ice_map = Nansat.from_domain(domain=raw_nansat,
array=classImage.astype(np.uint8))
ice_map.set_metadata(raw_nansat.get_metadata())
ice_map.set_metadata('entry_title', 'S1_SAR_ICE_MAP')
ice_map = add_colortable(ice_map, source)
ice_map.export(out_filename, bands=[1], driver='GTiff')
if quicklook:
rgb = colorcode_array(classImage, source)
plt.imsave(out_filename.replace('.tif', '.png'), rgb)
return out_filename
def test_crop(self):
n1 = Nansat(self.test_file_gcps, log_level=40, mapper=self.default_mapper)
ext = n1.crop(10, 20, 50, 60)
self.assertEqual(n1.shape(), (60, 50))
self.assertEqual(ext, (10, 20, 50, 60))
self.assertEqual(type(n1[1]), np.ndarray)
n1 = Nansat(self.test_file_gcps, log_level=40, mapper=self.default_mapper)
ext = n1.crop(0, 0, 200, 200)
self.assertEqual(n1.shape(), (200, 200))
self.assertEqual(ext, (0, 0, 200, 200))
self.assertEqual(type(n1[1]), np.ndarray)
def test_crop_no_gcps_arctic(self):
n1 = Nansat(self.test_file_arctic, logLevel=40)
ext = n1.crop(10, 20, 50, 60)
self.assertEqual(n1.shape(), (60, 50))
self.assertEqual(ext, (10, 20, 50, 60))
self.assertEqual(type(n1[1]), np.ndarray)
def test_crop_no_gcps_arctic(self):
n1 = Nansat(self.test_file_arctic, logLevel=40)
ext = n1.crop(10, 20, 50, 60)
self.assertEqual(n1.shape(), (60, 50))
self.assertEqual(ext, (10, 20, 50, 60))
self.assertEqual(type(n1[1]), np.ndarray)
def test_crop_lonlat_lims(self):
n1 = Nansat(self.test_file_gcps, logLevel=40)
st, ext = n1.crop(lonlim=[28, 29], latlim=[70.5, 71])
self.assertEqual(st, 0)
self.assertEqual(n1.shape(), (111, 110))
self.assertEqual(ext, (31, 89, 110, 111))
self.assertEqual(type(n1[1]), np.ndarray)
def test_crop_lonlat_lims(self):
n1 = Nansat(self.test_file_gcps, logLevel=40)
st, ext = n1.crop(lonlim=[28, 29], latlim=[70.5, 71])
self.assertEqual(st, 0)
self.assertEqual(n1.shape(), (111, 110))
self.assertEqual(ext, (31, 89, 110, 111))
self.assertEqual(type(n1[1]), np.ndarray)
def test_crop_gcpproj(self):
n1 = Nansat(self.test_file_gcps, log_level=40, mapper=self.default_mapper)
n1.reproject_gcps()
ext = n1.crop(10, 20, 50, 60)
xmed = abs(np.median(np.array([gcp.GCPX
for gcp in n1.vrt.dataset.GetGCPs()])))
gcpproj = NSR(n1.vrt.dataset.GetGCPProjection()
).ExportToProj4().split(' ')[0]
self.assertTrue(xmed > 360)
self.assertTrue(gcpproj=='+proj=stere')
def test_crop_gcpproj(self):
n1 = Nansat(self.test_file_gcps, logLevel=40)
n1.reproject_GCPs()
ext = n1.crop(10, 20, 50, 60)
xmed = abs(np.median(np.array([gcp.GCPX
for gcp in n1.vrt.dataset.GetGCPs()])))
gcpproj = NSR(n1.vrt.dataset.GetGCPProjection()
).ExportToProj4().split(' ')[0]
self.assertTrue(xmed > 360)
self.assertTrue(gcpproj=='+proj=stere')
kmeans = KMeans(n_clusters=9, n_jobs=cfg.numberOfThreads).fit(
pca.transform(scaler.transform(features_all))[:, :n_components])
pickle.dump([scaler, pca, n_components, kmeans], open(cfg.kmeansFilename,
"wb"))
# apply clustering
print('*** Exporting files to:')
for li, ifile in enumerate(ifiles):
ofile = ifile.replace('_texture_features.npz', '_kmeans_clustered.tif')
print('[%d/%d] %s' %
(li + 1, len(ifiles),
ifile.replace('_texture_features.npz', '_kmeans.tif')))
npz = np.load(ifile)
tfsHH = npz['textureFeatures'].item()['HH']
tfsHV = npz['textureFeatures'].item()['HV']
imgSize = tfsHH.shape[1:]
features = np.vstack([tfsHH, tfsHV]).reshape(26, np.prod(imgSize)).T
gpi = np.isfinite(features.sum(axis=1))
kmeansZones = np.zeros(np.prod(imgSize)) # 0 is reserved for void cells
kmeansZones[gpi] = 1 + kmeans.predict(
pca.transform(scaler.transform(features[gpi]))[:, :n_components])
kmeansZones = kmeansZones.reshape(imgSize)
nansatObjSigma0 = Nansat(
ifile.replace('_texture_features.npz', '_sigma0_HH_denoised.tif'))
if nansatObjSigma0.shape() != imgSize:
nansatObjSigma0.crop(0, 0, imgSize[1], imgSize[0])
nansatObjCluster = Nansat.from_domain(array=kmeansZones.astype(np.uint8),
domain=nansatObjSigma0)
nansatObjCluster.export(ofile, bands=[1], driver='GTiff')
if cfg.quicklook:
plt.imsave(ofile.replace('.tif', '.png'), kmeansZones, cmap='tab10')
pca = PCA(n_components=n_components).fit(scaler.transform(features_all))
kmeans = KMeans(n_clusters=n_clusters, n_jobs=cfg.numberOfThreads).fit(
pca.transform(scaler.transform(features_all)))
pickle.dump([scaler, pca, kmeans], open(cfg.kmeansFilename, "wb"))
# apply clustering
print('*** Exporting files to:')
for li, ifile in enumerate(ifiles):
ofile = ifile.replace('_texture_features.npz', '_kmeans_clustered.tif')
print('[%d/%d] %s' %
(li + 1, len(ifiles),
ifile.replace('_texture_features.npz', '_kmeans.tif')))
npz = np.load(ifile)
tfsHH = npz['textureFeatures'].item()['HH']
tfsHV = npz['textureFeatures'].item()['HV']
incAng = npz['incidenceAngle'][np.newaxis, :, :]
imgSize = tfsHH.shape[1:]
features = np.vstack([tfsHH, tfsHV, incAng]).reshape(27,
np.prod(imgSize)).T
gpi = np.isfinite(features.sum(axis=1))
kmeansZones = np.ones(np.prod(imgSize)) * np.nan
kmeansZones[gpi] = kmeans.predict(
pca.transform(scaler.transform(features[gpi])))
kmeansZones = kmeansZones.reshape(imgSize)
nansatObjGamma0 = Nansat(
ifile.replace('_texture_features.npz', '_denoised_gamma0_HH.tif'))
if nansatObjGamma0.shape() != imgSize:
nansatObjGamma0.crop(0, 0, imgSize[1], imgSize[0])
nansatObjCluster = Nansat(array=kmeansZones, domain=nansatObjGamma0)
nansatObjCluster.export(ofile, bands=[1], driver='GTiff')
plt.imsave(ofile.replace('.tif', '.png'), kmeansZones)
