def transform(self, X, y=None):
import numpy as np
from sklearn.decomposition import PCA
import pysptools.noise as ns
X = X.astype('float32')
# apply brightness normalization
# if raster
if self.BrightnessNormalization == True:
def norm(r):
norm = r / np.sqrt(np.sum((r**2), 0))
return norm
if len(X.shape) == 3:
X = np.apply_along_axis(norm, 2, X)
# if 2D array
if len(X.shape) == 2:
X = np.apply_along_axis(norm, 0, X)
w = ns.Whiten()
wdata = w.apply(X)
numBands = X.shape[2]
h, w, numBands = wdata.shape
X = np.reshape(wdata, (w * h, numBands))
pca = PCA()
mnf = pca.fit_transform(X)
mnf = np.reshape(mnf, (h, w, numBands))
mnf = mnf[:, :, :self.n_components]
var = np.cumsum(
np.round(pca.explained_variance_ratio_, decimals=4) * 100)
return mnf, var
def explained_variance(img):
from sklearn.decomposition import PCA
w = ns.Whiten()
wdata = w.apply(img)
numBands = r.count
h, w, numBands = wdata.shape
X = np.reshape(wdata, (w * h, numBands))
pca = PCA()
mnf = pca.fit_transform(X)
return print(
np.cumsum(np.round(pca.explained_variance_ratio_, decimals=4) * 100))
def whiten(data):
w = ns.Whiten()
return w.apply(data)
explained_variance(img)
else:
if args['method'] == 1:
# Load raster/convert to ndarray format
name = os.path.basename(inImage)
r = rasterio.open(inImage)
r2 = r.read()
# Apply Brightness Normalization if the option -p is added
if args["preprop"] == True:
r2 = np.apply_along_axis(BrigthnessNormalization, 0, r2)
img = reshape_as_image(r2)
# Apply MNF -m 1
print("Creating MNF components of " + name)
# Apply MNF namualy acording to pysptools
w = ns.Whiten()
wdata = w.apply(img)
numBands = r.count
h, w, numBands = wdata.shape
X = np.reshape(wdata, (w * h, numBands))
pca = PCA()
mnf = pca.fit_transform(X)
mnf = np.reshape(mnf, (h, w, numBands))
if args["SavitzkyGolay"] == True:
dn = ns.SavitzkyGolay()
mnf[:, :, 1:2] = dn.denoise_bands(mnf[:, :, 1:2], 15, 2)
mnf = mnf[:, :, :n_components]
saveMNF(mnf, r)
# Apply MNF coefficients to the other images
for i in range(len(imageList)):