forked from kilianv/CoastSat_WRL
Neural Network
kvos
7 years ago
parent
0af37d9c6a
commit
d2c344ec99
File diff suppressed because it is too large
Load Diff
@ -0,0 +1,685 @@
|
||||
# -*- coding: utf-8 -*-
|
||||
"""
|
||||
Created on Thu Mar 1 11:20:35 2018
|
||||
|
||||
@author: z5030440
|
||||
"""
|
||||
|
||||
"""This script contains the functions needed for satellite derived shoreline (SDS) extraction"""
|
||||
|
||||
# Initial settings
|
||||
import numpy as np
|
||||
import matplotlib.pyplot as plt
|
||||
import pdb
|
||||
import ee
|
||||
|
||||
# other modules
|
||||
from osgeo import gdal, ogr, osr
|
||||
import tempfile
|
||||
from urllib.request import urlretrieve
|
||||
import zipfile
|
||||
|
||||
# image processing modules
|
||||
import skimage.filters as filters
|
||||
import skimage.exposure as exposure
|
||||
import skimage.transform as transform
|
||||
import sklearn.decomposition as decomposition
|
||||
import skimage.measure as measure
|
||||
import skimage.morphology as morphology
|
||||
from sklearn.cluster import KMeans
|
||||
|
||||
|
||||
# import own modules
|
||||
from functions.utils import *
|
||||
|
||||
|
||||
# Download from ee server function
|
||||
|
||||
def download_tif(image, polygon, bandsId):
|
||||
"""downloads tif image (region and bands) from the ee server and stores it in a temp file"""
|
||||
url = ee.data.makeDownloadUrl(ee.data.getDownloadId({
|
||||
'image': image.serialize(),
|
||||
'region': polygon,
|
||||
'bands': bandsId,
|
||||
'filePerBand': 'false',
|
||||
'name': 'data',
|
||||
}))
|
||||
local_zip, headers = urlretrieve(url)
|
||||
with zipfile.ZipFile(local_zip) as local_zipfile:
|
||||
return local_zipfile.extract('data.tif', tempfile.mkdtemp())
|
||||
|
||||
def load_image(image, polygon, bandsId):
|
||||
"""
|
||||
Loads an ee.Image() as a np.array. e.Image() is retrieved from the EE database.
|
||||
The geographic area and bands to select can be specified
|
||||
|
||||
KV WRL 2018
|
||||
|
||||
Arguments:
|
||||
-----------
|
||||
image: ee.Image()
|
||||
image objec from the EE database
|
||||
polygon: list
|
||||
coordinates of the points creating a polygon. Each point is a list with 2 values
|
||||
bandsId: list
|
||||
bands to select, each band is a dictionnary in the list containing the following keys:
|
||||
crs, crs_transform, data_type and id. NOTE: you have to remove the key dimensions, otherwise
|
||||
the entire image is retrieved.
|
||||
|
||||
Returns:
|
||||
-----------
|
||||
image_array : np.ndarray
|
||||
An array containing the image (2D if one band, otherwise 3D)
|
||||
georef : np.ndarray
|
||||
6 element vector containing the crs_parameters
|
||||
[X_ul_corner Xscale Xshear Y_ul_corner Yshear Yscale]
|
||||
"""
|
||||
|
||||
local_tif_filename = download_tif(image, polygon, bandsId)
|
||||
dataset = gdal.Open(local_tif_filename, gdal.GA_ReadOnly)
|
||||
georef = np.array(dataset.GetGeoTransform())
|
||||
bands = [dataset.GetRasterBand(i + 1).ReadAsArray() for i in range(dataset.RasterCount)]
|
||||
return np.stack(bands, 2), georef
|
||||
|
||||
def create_cloud_mask(im_qa, satname, plot_bool):
|
||||
"""
|
||||
Creates a cloud mask from the image containing the QA band information
|
||||
|
||||
KV WRL 2018
|
||||
|
||||
Arguments:
|
||||
-----------
|
||||
im_qa: np.ndarray
|
||||
Image containing the QA band
|
||||
satname: string
|
||||
short name for the satellite (L8, L7, S2)
|
||||
plot_bool: boolean
|
||||
True if plot is wanted
|
||||
|
||||
Returns:
|
||||
-----------
|
||||
cloud_mask : np.ndarray of booleans
|
||||
A boolean array with True where the cloud are present
|
||||
"""
|
||||
|
||||
# convert QA bits
|
||||
if satname == 'L8':
|
||||
cloud_values = [2800, 2804, 2808, 2812, 6896, 6900, 6904, 6908]
|
||||
elif satname == 'L7':
|
||||
cloud_values = [752, 756, 760, 764]
|
||||
|
||||
cloud_mask = np.isin(im_qa, cloud_values)
|
||||
# remove isolated cloud pixels (there are some in the swash and they cause problems)
|
||||
if sum(sum(cloud_mask)) > 0:
|
||||
morphology.remove_small_objects(cloud_mask, min_size=10, connectivity=1, in_place=True)
|
||||
|
||||
if plot_bool:
|
||||
plt.figure()
|
||||
plt.imshow(cloud_mask, cmap='gray')
|
||||
plt.draw()
|
||||
|
||||
#cloud_shadow_values = [2976, 2980, 2984, 2988, 3008, 3012, 3016, 3020]
|
||||
#cloud_shadow_mask = np.isin(im_qa, cloud_shadow_values)
|
||||
|
||||
return cloud_mask
|
||||
|
||||
def read_eeimage(im, polygon, sat_name, plot_bool):
|
||||
"""
|
||||
Read an ee.Image() object and returns the panchromatic band, multispectral bands (B, G, R, NIR, SWIR)
|
||||
and a cloud mask. All outputs are at 15m resolution (bilinear interpolation for the multispectral bands)
|
||||
|
||||
KV WRL 2018
|
||||
|
||||
Arguments:
|
||||
-----------
|
||||
im: ee.Image()
|
||||
Image to read from the Google Earth Engine database
|
||||
plot_bool: boolean
|
||||
True if plot is wanted
|
||||
|
||||
Returns:
|
||||
-----------
|
||||
im_pan: np.ndarray (2D)
|
||||
The panchromatic band (15m)
|
||||
im_ms: np.ndarray (3D)
|
||||
The multispectral bands interpolated at 15m
|
||||
im_cloud: np.ndarray (2D)
|
||||
The cloud mask at 15m
|
||||
crs_params: list
|
||||
EPSG code and affine transformation parameters
|
||||
"""
|
||||
|
||||
im_dic = im.getInfo()
|
||||
# save metadata
|
||||
im_meta = im_dic.get('properties')
|
||||
meta = {'timestamp':im_meta['system:time_start'],
|
||||
'date_acquired':im_meta['DATE_ACQUIRED'],
|
||||
'geom_rmse_model':im_meta['GEOMETRIC_RMSE_MODEL'],
|
||||
'gcp_model':im_meta['GROUND_CONTROL_POINTS_MODEL'],
|
||||
'quality':im_meta['IMAGE_QUALITY_OLI'],
|
||||
'sun_azimuth':im_meta['SUN_AZIMUTH'],
|
||||
'sun_elevation':im_meta['SUN_ELEVATION']}
|
||||
|
||||
im_bands = im_dic.get('bands')
|
||||
|
||||
# delete dimensions key from dictionnary, otherwise the entire image is extracted
|
||||
for i in range(len(im_bands)): del im_bands[i]['dimensions']
|
||||
|
||||
# load panchromatic band
|
||||
pan_band = [im_bands[7]]
|
||||
im_pan, crs_pan = load_image(im, polygon, pan_band)
|
||||
im_pan = im_pan[:,:,0]
|
||||
|
||||
# load the multispectral bands (B2,B3,B4,B5,B6) = (blue,green,red,nir,swir1)
|
||||
ms_bands = [im_bands[1], im_bands[2], im_bands[3], im_bands[4], im_bands[5]]
|
||||
im_ms_30m, crs_ms = load_image(im, polygon, ms_bands)
|
||||
|
||||
# create cloud mask
|
||||
qa_band = [im_bands[11]]
|
||||
im_qa, crs_qa = load_image(im, polygon, qa_band)
|
||||
im_qa = im_qa[:,:,0]
|
||||
im_cloud = create_cloud_mask(im_qa, sat_name, plot_bool)
|
||||
im_cloud = transform.resize(im_cloud, (im_pan.shape[0], im_pan.shape[1]),
|
||||
order=0, preserve_range=True, mode='constant').astype('bool_')
|
||||
|
||||
# resize the image using bilinear interpolation (order 1)
|
||||
im_ms = transform.resize(im_ms_30m,(im_pan.shape[0], im_pan.shape[1]),
|
||||
order=1, preserve_range=True, mode='constant')
|
||||
|
||||
# check if -inf values (means out of image) and add to cloud mask
|
||||
im_inf = np.isin(im_ms[:,:,0], -np.inf)
|
||||
im_nan = np.isnan(im_ms[:,:,0])
|
||||
im_cloud = np.logical_or(np.logical_or(im_cloud, im_inf), im_nan)
|
||||
|
||||
# get the crs parameters for the image at 15m and 30m resolution
|
||||
crs = {'crs_15m':crs_pan, 'crs_30m':crs_ms, 'epsg_code':int(pan_band[0]['crs'][5:])}
|
||||
|
||||
if plot_bool:
|
||||
|
||||
# if there are -inf in the image, set them to 0 before plotting
|
||||
if sum(sum(np.isin(im_ms_30m[:,:,0], -np.inf).astype(int))) > 0:
|
||||
idx = np.isin(im_ms_30m[:,:,0], -np.inf)
|
||||
im_ms_30m[idx,0] = 0; im_ms_30m[idx,1] = 0; im_ms_30m[idx,2] = 0;
|
||||
im_ms_30m[idx,3] = 0; im_ms_30m[idx,4] = 0
|
||||
|
||||
plt.figure()
|
||||
|
||||
plt.subplot(221)
|
||||
plt.imshow(im_pan, cmap='gray')
|
||||
plt.title('PANCHROMATIC')
|
||||
|
||||
plt.subplot(222)
|
||||
plt.imshow(im_ms_30m[:,:,[2,1,0]])
|
||||
plt.title('RGB')
|
||||
|
||||
|
||||
plt.subplot(223)
|
||||
plt.imshow(im_ms_30m[:,:,3], cmap='gray')
|
||||
plt.title('NIR')
|
||||
|
||||
plt.subplot(224)
|
||||
plt.imshow(im_ms_30m[:,:,4], cmap='gray')
|
||||
plt.title('SWIR')
|
||||
|
||||
plt.show()
|
||||
|
||||
return im_pan, im_ms, im_cloud, crs, meta
|
||||
|
||||
|
||||
def rescale_image_intensity(im, cloud_mask, prob_high, plot_bool):
|
||||
"""
|
||||
Rescales the intensity of an image (multispectral or single band) by applying
|
||||
a cloud mask and clipping the prob_high upper percentile. This functions allows
|
||||
to stretch the contrast of an image.
|
||||
|
||||
KV WRL 2018
|
||||
|
||||
Arguments:
|
||||
-----------
|
||||
im: np.ndarray
|
||||
Image to rescale, can be 3D (multispectral) or 2D (single band)
|
||||
cloud_mask: np.ndarray
|
||||
2D cloud mask with True where cloud pixels are
|
||||
prob_high: float
|
||||
probability of exceedence used to calculate the upper percentile
|
||||
plot_bool: boolean
|
||||
True if plot is wanted
|
||||
|
||||
Returns:
|
||||
-----------
|
||||
im_adj: np.ndarray
|
||||
The rescaled image
|
||||
"""
|
||||
prc_low = 0 # lower percentile
|
||||
vec_mask = cloud_mask.reshape(im.shape[0] * im.shape[1])
|
||||
|
||||
if plot_bool:
|
||||
plt.figure()
|
||||
|
||||
if len(im.shape) > 2:
|
||||
vec = im.reshape(im.shape[0] * im.shape[1], im.shape[2])
|
||||
vec_adj = np.ones((len(vec_mask), im.shape[2])) * np.nan
|
||||
|
||||
for i in range(im.shape[2]):
|
||||
prc_high = np.percentile(vec[~vec_mask, i], prob_high)
|
||||
vec_rescaled = exposure.rescale_intensity(vec[~vec_mask, i], in_range=(prc_low, prc_high))
|
||||
vec_adj[~vec_mask,i] = vec_rescaled
|
||||
|
||||
if plot_bool:
|
||||
plt.subplot(np.floor(im.shape[2]/2) + 1, np.floor(im.shape[2]/2), i+1)
|
||||
plt.hist(vec[~vec_mask, i], bins=200, label='original')
|
||||
plt.hist(vec_rescaled, bins=200, alpha=0.5, label='rescaled')
|
||||
plt.legend()
|
||||
plt.title('Band' + str(i+1))
|
||||
plt.show()
|
||||
|
||||
im_adj = vec_adj.reshape(im.shape[0], im.shape[1], im.shape[2])
|
||||
|
||||
if plot_bool:
|
||||
plt.figure()
|
||||
ax1 = plt.subplot(121)
|
||||
plt.imshow(im[:,:,[2,1,0]])
|
||||
plt.axis('off')
|
||||
plt.title('Original')
|
||||
ax2 = plt.subplot(122, sharex=ax1, sharey=ax1)
|
||||
plt.imshow(im_adj[:,:,[2,1,0]])
|
||||
plt.axis('off')
|
||||
plt.title('Rescaled')
|
||||
plt.show()
|
||||
|
||||
else:
|
||||
vec = im.reshape(im.shape[0] * im.shape[1])
|
||||
vec_adj = np.ones(len(vec_mask)) * np.nan
|
||||
prc_high = np.percentile(vec[~vec_mask], prob_high)
|
||||
vec_rescaled = exposure.rescale_intensity(vec[~vec_mask], in_range=(prc_low, prc_high))
|
||||
vec_adj[~vec_mask] = vec_rescaled
|
||||
|
||||
if plot_bool:
|
||||
plt.hist(vec[~vec_mask], bins=200, label='original')
|
||||
plt.hist(vec_rescaled, bins=200, alpha=0.5, label='rescaled')
|
||||
plt.legend()
|
||||
plt.title('Single band')
|
||||
plt.show()
|
||||
|
||||
im_adj = vec_adj.reshape(im.shape[0], im.shape[1])
|
||||
|
||||
if plot_bool:
|
||||
plt.figure()
|
||||
ax1 = plt.subplot(121)
|
||||
plt.imshow(im, cmap='gray')
|
||||
plt.axis('off')
|
||||
plt.title('Original')
|
||||
ax2 = plt.subplot(122, sharex=ax1, sharey=ax1)
|
||||
plt.imshow(im_adj, cmap='gray')
|
||||
plt.axis('off')
|
||||
plt.title('Rescaled')
|
||||
plt.show()
|
||||
|
||||
return im_adj
|
||||
|
||||
|
||||
def hist_match(source, template):
|
||||
"""
|
||||
Adjust the pixel values of a grayscale image such that its histogram
|
||||
matches that of a target image
|
||||
|
||||
Arguments:
|
||||
-----------
|
||||
source: np.ndarray
|
||||
Image to transform; the histogram is computed over the flattened
|
||||
array
|
||||
template: np.ndarray
|
||||
Template image; can have different dimensions to source
|
||||
Returns:
|
||||
-----------
|
||||
matched: np.ndarray
|
||||
The transformed output image
|
||||
"""
|
||||
|
||||
oldshape = source.shape
|
||||
source = source.ravel()
|
||||
template = template.ravel()
|
||||
|
||||
# get the set of unique pixel values and their corresponding indices and
|
||||
# counts
|
||||
s_values, bin_idx, s_counts = np.unique(source, return_inverse=True,
|
||||
return_counts=True)
|
||||
t_values, t_counts = np.unique(template, return_counts=True)
|
||||
|
||||
# take the cumsum of the counts and normalize by the number of pixels to
|
||||
# get the empirical cumulative distribution functions for the source and
|
||||
# template images (maps pixel value --> quantile)
|
||||
s_quantiles = np.cumsum(s_counts).astype(np.float64)
|
||||
s_quantiles /= s_quantiles[-1]
|
||||
t_quantiles = np.cumsum(t_counts).astype(np.float64)
|
||||
t_quantiles /= t_quantiles[-1]
|
||||
|
||||
# interpolate linearly to find the pixel values in the template image
|
||||
# that correspond most closely to the quantiles in the source image
|
||||
interp_t_values = np.interp(s_quantiles, t_quantiles, t_values)
|
||||
|
||||
return interp_t_values[bin_idx].reshape(oldshape)
|
||||
|
||||
def pansharpen(im_ms, im_pan, cloud_mask, plot_bool):
|
||||
"""
|
||||
Pansharpens a multispectral image (3D), using the panchromatic band (2D)
|
||||
and a cloud mask
|
||||
|
||||
KV WRL 2018
|
||||
|
||||
Arguments:
|
||||
-----------
|
||||
im_ms: np.ndarray
|
||||
Multispectral image to pansharpen (3D)
|
||||
im_pan: np.ndarray
|
||||
Panchromatic band (2D)
|
||||
cloud_mask: np.ndarray
|
||||
2D cloud mask with True where cloud pixels are
|
||||
plot_bool: boolean
|
||||
True if plot is wanted
|
||||
|
||||
Returns:
|
||||
-----------
|
||||
im_ms_ps: np.ndarray
|
||||
Pansharpened multisoectral image (3D)
|
||||
"""
|
||||
|
||||
# reshape image into vector and apply cloud mask
|
||||
vec = im_ms.reshape(im_ms.shape[0] * im_ms.shape[1], im_ms.shape[2])
|
||||
vec_mask = cloud_mask.reshape(im_ms.shape[0] * im_ms.shape[1])
|
||||
vec = vec[~vec_mask, :]
|
||||
# apply PCA to RGB bands
|
||||
pca = decomposition.PCA()
|
||||
vec_pcs = pca.fit_transform(vec)
|
||||
# replace 1st PC with pan band (after matching histograms)
|
||||
vec_pan = im_pan.reshape(im_pan.shape[0] * im_pan.shape[1])
|
||||
vec_pan = vec_pan[~vec_mask]
|
||||
vec_pcs[:,0] = hist_match(vec_pan, vec_pcs[:,0])
|
||||
vec_ms_ps = pca.inverse_transform(vec_pcs)
|
||||
|
||||
# normalise between 0 and 1
|
||||
for i in range(vec_pcs.shape[1]):
|
||||
vec_ms_ps[:,i] = np.divide(vec_ms_ps[:,i] - np.min(vec_ms_ps[:,i]),
|
||||
np.max(vec_ms_ps[:,i]) - np.min(vec_ms_ps[:,i]))
|
||||
# reshape vector into image
|
||||
vec_ms_ps_full = np.ones((len(vec_mask), im_ms.shape[2])) * np.nan
|
||||
vec_ms_ps_full[~vec_mask,:] = vec_ms_ps
|
||||
im_ms_ps = vec_ms_ps_full.reshape(im_ms.shape[0], im_ms.shape[1], im_ms.shape[2])
|
||||
|
||||
if plot_bool:
|
||||
plt.figure()
|
||||
ax1 = plt.subplot(121)
|
||||
plt.imshow(im_ms[:,:,[2,1,0]])
|
||||
plt.axis('off')
|
||||
plt.title('Original')
|
||||
ax2 = plt.subplot(122, sharex=ax1, sharey=ax1)
|
||||
plt.imshow(im_ms_ps[:,:,[2,1,0]])
|
||||
plt.axis('off')
|
||||
plt.title('Pansharpened')
|
||||
plt.show()
|
||||
|
||||
return im_ms_ps
|
||||
|
||||
def nd_index(im1, im2, cloud_mask, plot_bool):
|
||||
"""
|
||||
Computes normalised difference index on 2 images (2D), given a cloud mask (2D)
|
||||
|
||||
KV WRL 2018
|
||||
|
||||
Arguments:
|
||||
-----------
|
||||
im1, im2: np.ndarray
|
||||
Images (2D) with which to calculate the ND index
|
||||
cloud_mask: np.ndarray
|
||||
2D cloud mask with True where cloud pixels are
|
||||
plot_bool: boolean
|
||||
True if plot is wanted
|
||||
|
||||
Returns: -----------
|
||||
im_nd: np.ndarray
|
||||
|
||||
Image (2D) containing the ND index
|
||||
"""
|
||||
vec_mask = cloud_mask.reshape(im1.shape[0] * im1.shape[1])
|
||||
vec_nd = np.ones(len(vec_mask)) * np.nan
|
||||
vec1 = im1.reshape(im1.shape[0] * im1.shape[1])
|
||||
vec2 = im2.reshape(im2.shape[0] * im2.shape[1])
|
||||
temp = np.divide(vec1[~vec_mask] - vec2[~vec_mask],
|
||||
vec1[~vec_mask] + vec2[~vec_mask])
|
||||
vec_nd[~vec_mask] = temp
|
||||
im_nd = vec_nd.reshape(im1.shape[0], im1.shape[1])
|
||||
|
||||
if plot_bool:
|
||||
plt.figure()
|
||||
plt.imshow(im_nd, cmap='seismic')
|
||||
plt.colorbar()
|
||||
plt.title('Normalised index')
|
||||
plt.show()
|
||||
|
||||
return im_nd
|
||||
|
||||
def find_wl_contours(im_ndwi, cloud_mask, min_contour_points, plot_bool):
|
||||
"""
|
||||
Computes normalised difference index on 2 images (2D), given a cloud mask (2D)
|
||||
|
||||
KV WRL 2018
|
||||
|
||||
Arguments:
|
||||
-----------
|
||||
im_ndwi: np.ndarray
|
||||
Image (2D) with the NDWI (water index)
|
||||
cloud_mask: np.ndarray
|
||||
2D cloud mask with True where cloud pixels are
|
||||
min_contour_points: int
|
||||
minimum number of points in each contour line
|
||||
plot_bool: boolean
|
||||
True if plot is wanted
|
||||
|
||||
Returns: -----------
|
||||
contours_wl: list of np.arrays
|
||||
contains the (row,column) coordinates of the contour lines
|
||||
|
||||
"""
|
||||
|
||||
# reshape image to vector
|
||||
vec_ndwi = im_ndwi.reshape(im_ndwi.shape[0] * im_ndwi.shape[1])
|
||||
vec_mask = cloud_mask.reshape(cloud_mask.shape[0] * cloud_mask.shape[1])
|
||||
vec = vec_ndwi[~vec_mask]
|
||||
# apply otsu's threshold
|
||||
t_otsu = filters.threshold_otsu(vec)
|
||||
# use Marching Squares algorithm to detect contours on ndwi image
|
||||
contours = measure.find_contours(im_ndwi, t_otsu)
|
||||
# filter water lines
|
||||
contours_wl = []
|
||||
for i, contour in enumerate(contours):
|
||||
# remove contour points that are around clouds (nan values)
|
||||
if np.any(np.isnan(contour)):
|
||||
index_nan = np.where(np.isnan(contour))[0]
|
||||
contour = np.delete(contour, index_nan, axis=0)
|
||||
# remove contours that have only few points (less than min_contour_points)
|
||||
if contour.shape[0] > min_contour_points:
|
||||
contours_wl.append(contour)
|
||||
|
||||
if plot_bool:
|
||||
# plot otsu's histogram segmentation
|
||||
plt.figure()
|
||||
vals = plt.hist(vec, bins=200)
|
||||
plt.plot([t_otsu, t_otsu],[0, np.max(vals[0])], 'r-', label='Otsu threshold')
|
||||
plt.legend()
|
||||
plt.show()
|
||||
|
||||
# plot the water line contours on top of water index
|
||||
plt.figure()
|
||||
plt.imshow(im_ndwi, cmap='seismic')
|
||||
plt.colorbar()
|
||||
for i,contour in enumerate(contours_wl): plt.plot(contour[:, 1], contour[:, 0], linewidth=3, color='k')
|
||||
plt.axis('image')
|
||||
plt.title('Detected water lines')
|
||||
plt.show()
|
||||
|
||||
return contours_wl
|
||||
|
||||
def convert_pix2world(points, crs_vec):
|
||||
"""
|
||||
Converts pixel coordinates (row,columns) to world projected coordinates
|
||||
performing an affine transformation
|
||||
|
||||
KV WRL 2018
|
||||
|
||||
Arguments:
|
||||
-----------
|
||||
points: np.ndarray or list of np.ndarray
|
||||
array with 2 columns (rows first and columns second)
|
||||
crs_vec: np.ndarray
|
||||
vector of 6 elements [Xtr, Xscale, Xshear, Ytr, Yshear, Yscale]
|
||||
|
||||
Returns: -----------
|
||||
points_converted: np.ndarray or list of np.ndarray
|
||||
converted coordinates, first columns with X and second column with Y
|
||||
|
||||
"""
|
||||
|
||||
# make affine transformation matrix
|
||||
aff_mat = np.array([[crs_vec[1], crs_vec[2], crs_vec[0]],
|
||||
[crs_vec[4], crs_vec[5], crs_vec[3]],
|
||||
[0, 0, 1]])
|
||||
# create affine transformation
|
||||
tform = transform.AffineTransform(aff_mat)
|
||||
|
||||
if type(points) is list:
|
||||
points_converted = []
|
||||
# iterate over the list
|
||||
for i, arr in enumerate(points):
|
||||
tmp = arr[:,[1,0]]
|
||||
points_converted.append(tform(tmp))
|
||||
|
||||
elif type(points) is np.ndarray:
|
||||
tmp = points[:,[1,0]]
|
||||
points_converted = tform(tmp)
|
||||
|
||||
else:
|
||||
print('invalid input type')
|
||||
raise
|
||||
|
||||
return points_converted
|
||||
|
||||
def convert_epsg(points, epsg_in, epsg_out):
|
||||
"""
|
||||
Converts from one spatial reference to another using the epsg codes
|
||||
|
||||
KV WRL 2018
|
||||
|
||||
Arguments:
|
||||
-----------
|
||||
points: np.ndarray or list of np.ndarray
|
||||
array with 2 columns (rows first and columns second)
|
||||
epsg_in: int
|
||||
epsg code of the spatial reference in which the input is
|
||||
epsg_out: int
|
||||
epsg code of the spatial reference in which the output will be
|
||||
|
||||
Returns: -----------
|
||||
points_converted: np.ndarray or list of np.ndarray
|
||||
converted coordinates
|
||||
|
||||
"""
|
||||
|
||||
# define input and output spatial references
|
||||
inSpatialRef = osr.SpatialReference()
|
||||
inSpatialRef.ImportFromEPSG(epsg_in)
|
||||
outSpatialRef = osr.SpatialReference()
|
||||
outSpatialRef.ImportFromEPSG(epsg_out)
|
||||
# create a coordinates transform
|
||||
coordTransform = osr.CoordinateTransformation(inSpatialRef, outSpatialRef)
|
||||
# transform points
|
||||
if type(points) is list:
|
||||
points_converted = []
|
||||
# iterate over the list
|
||||
for i, arr in enumerate(points):
|
||||
points_converted.append(np.array(coordTransform.TransformPoints(arr)))
|
||||
|
||||
elif type(points) is np.ndarray:
|
||||
points_converted = np.array(coordTransform.TransformPoints(points))
|
||||
|
||||
else:
|
||||
print('invalid input type')
|
||||
raise
|
||||
|
||||
return points_converted
|
||||
|
||||
def classify_sand_unsupervised(im_ms_ps, im_pan, cloud_mask, wl_pix, buffer_size, min_beach_size, plot_bool):
|
||||
"""
|
||||
Classifies sand pixels using an unsupervised algorithm (Kmeans)
|
||||
Set buffer size to False if you
|
||||
|
||||
KV WRL 2018
|
||||
|
||||
Arguments:
|
||||
-----------
|
||||
im_ms_ps: np.ndarray
|
||||
Pansharpened RGB + downsampled NIR and SWIR
|
||||
im_pan:
|
||||
Panchromatic band
|
||||
cloud_mask: np.ndarray
|
||||
2D cloud mask with True where cloud pixels are
|
||||
wl_pix: list of np.ndarray
|
||||
list of arrays containig the pixel coordinates of the water line
|
||||
buffer_size: int or False
|
||||
radius of the disk used to create a buffer around the water line
|
||||
when False, the entire image is considered for kmeans
|
||||
min_beach_size: int
|
||||
minimum number of connected pixels belonging to a single beach
|
||||
plot_bool: boolean
|
||||
True if plot is wanted
|
||||
|
||||
Returns: -----------
|
||||
im_sand: np.ndarray
|
||||
2D binary image containing True where sand pixels are located
|
||||
|
||||
"""
|
||||
# reshape the 2D images into vectors
|
||||
vec_ms_ps = im_ms_ps.reshape(im_ms_ps.shape[0] * im_ms_ps.shape[1], im_ms_ps.shape[2])
|
||||
vec_pan = im_pan.reshape(im_pan.shape[0]*im_pan.shape[1])
|
||||
vec_mask = cloud_mask.reshape(im_ms_ps.shape[0] * im_ms_ps.shape[1])
|
||||
# add B,G,R,NIR and pan bands to the vector of features
|
||||
vec_features = np.zeros((vec_ms_ps.shape[0], 5))
|
||||
vec_features[:,[0,1,2,3]] = vec_ms_ps[:,[0,1,2,3]]
|
||||
vec_features[:,4] = vec_pan
|
||||
|
||||
if buffer_size:
|
||||
# create binary image with ones where the detected water lines is
|
||||
im_buffer = np.zeros((im_ms_ps.shape[0], im_ms_ps.shape[1]))
|
||||
for i, contour in enumerate(wl_pix):
|
||||
indices = [(int(_[0]), int(_[1])) for _ in list(np.round(contour))]
|
||||
for j, idx in enumerate(indices):
|
||||
im_buffer[idx] = 1
|
||||
# perform a dilation on the binary image
|
||||
se = morphology.disk(buffer_size)
|
||||
im_buffer = morphology.binary_dilation(im_buffer, se)
|
||||
vec_buffer = (im_buffer == 1).reshape(im_ms_ps.shape[0] * im_ms_ps.shape[1])
|
||||
else:
|
||||
vec_buffer = np.ones((vec_pan.shape[0]))
|
||||
# add cloud mask to buffer
|
||||
vec_buffer= np.logical_and(vec_buffer, ~vec_mask)
|
||||
# perform kmeans (6 clusters)
|
||||
kmeans = KMeans(n_clusters=6, random_state=0).fit(vec_features[vec_buffer,:])
|
||||
labels = np.ones((len(vec_mask))) * np.nan
|
||||
labels[vec_buffer] = kmeans.labels_
|
||||
im_labels = labels.reshape(im_ms_ps.shape[0], im_ms_ps.shape[1])
|
||||
# find the class with maximum reflection in the B,G,R,Pan
|
||||
im_sand = im_labels == np.argmax(np.mean(kmeans.cluster_centers_[:,[0,1,2,4]], axis=1))
|
||||
im_sand = morphology.remove_small_objects(im_sand, min_size=min_beach_size, connectivity=2)
|
||||
# im_sand = morphology.binary_dilation(im_sand, morphology.disk(1))
|
||||
|
||||
if plot_bool:
|
||||
im = np.copy(im_ms_ps)
|
||||
im[im_sand,0] = 0
|
||||
im[im_sand,1] = 0
|
||||
im[im_sand,2] = 1
|
||||
plt.figure()
|
||||
plt.imshow(im[:,:,[2,1,0]])
|
||||
plt.axis('image')
|
||||
plt.title('Sand classification')
|
||||
plt.show()
|
||||
|
||||
return im_sand
|
||||
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
Binary file not shown.
File diff suppressed because it is too large
Load Diff
@ -0,0 +1,96 @@
|
||||
# -*- coding: utf-8 -*-
|
||||
|
||||
#==========================================================#
|
||||
# Train Neural Network
|
||||
#==========================================================#
|
||||
|
||||
# Initial settings
|
||||
import os
|
||||
import numpy as np
|
||||
import matplotlib.pyplot as plt
|
||||
import ee
|
||||
import pdb
|
||||
import time
|
||||
import pandas as pd
|
||||
# other modules
|
||||
from osgeo import gdal, ogr, osr
|
||||
import pickle
|
||||
import matplotlib.cm as cm
|
||||
from pylab import ginput
|
||||
|
||||
# image processing modules
|
||||
import skimage.filters as filters
|
||||
import skimage.exposure as exposure
|
||||
import skimage.transform as transform
|
||||
import sklearn.decomposition as decomposition
|
||||
import skimage.measure as measure
|
||||
import skimage.morphology as morphology
|
||||
from scipy import ndimage
|
||||
import random
|
||||
|
||||
# machine learning modules
|
||||
from sklearn.model_selection import train_test_split
|
||||
from sklearn.neural_network import MLPClassifier
|
||||
from sklearn.preprocessing import StandardScaler, Normalizer
|
||||
from sklearn.externals import joblib
|
||||
|
||||
# import own modules
|
||||
import functions.utils as utils
|
||||
import functions.sds as sds
|
||||
|
||||
# some settings
|
||||
np.seterr(all='ignore') # raise/ignore divisions by 0 and nans
|
||||
plt.rcParams['axes.grid'] = True
|
||||
plt.rcParams['figure.max_open_warning'] = 100
|
||||
ee.Initialize()
|
||||
#%%
|
||||
|
||||
# load training data
|
||||
sitename = 'NARRA'
|
||||
with open(os.path.join(os.getcwd(), 'sand_classification', sitename + '_sand.pkl'), 'rb') as f:
|
||||
train_sand = pickle.load(f)
|
||||
with open(os.path.join(os.getcwd(), 'sand_classification', sitename + '_swash.pkl'), 'rb') as f:
|
||||
train_swash = pickle.load(f)
|
||||
with open(os.path.join(os.getcwd(), 'sand_classification', sitename + '_water.pkl'), 'rb') as f:
|
||||
train_water = pickle.load(f)
|
||||
with open(os.path.join(os.getcwd(), 'sand_classification', sitename + '_other.pkl'), 'rb') as f:
|
||||
train_other = pickle.load(f)
|
||||
|
||||
|
||||
train_water = train_water[np.random.choice(train_water.shape[0], 1500),:]
|
||||
train_other = train_other[np.random.choice(train_other.shape[0], 1500),:]
|
||||
|
||||
n_features = train_sand.shape[1]
|
||||
n_sand = len(train_sand)
|
||||
n_swash = len(train_swash)
|
||||
n_water = len(train_water)
|
||||
n_other = len(train_other)
|
||||
|
||||
training_data = np.zeros((n_sand+n_swash+n_water+n_other, n_features+1))
|
||||
training_data[:n_sand,:n_features] = train_sand
|
||||
training_data[n_sand:n_sand+n_swash,:n_features] = train_swash
|
||||
training_data[n_sand+n_swash:n_sand+n_swash+n_water,:n_features] = train_water
|
||||
training_data[n_sand+n_swash+n_water:n_sand+n_swash+n_water+n_other,:n_features] = train_other
|
||||
training_data[:n_sand,n_features] = 1*np.ones((n_sand))
|
||||
training_data[n_sand:n_sand+n_swash,n_features] = 2*np.ones((n_swash))
|
||||
training_data[n_sand+n_swash:n_sand+n_swash+n_water,n_features] = 3*np.ones((n_water))
|
||||
|
||||
X = training_data[:,0:n_features]
|
||||
y = training_data[:,n_features]
|
||||
|
||||
# divide in train and test
|
||||
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=0)
|
||||
|
||||
# standardize data
|
||||
#scaler = StandardScaler()
|
||||
#scaler.fit(X_train)
|
||||
#X_train = scaler.transform(X_train)
|
||||
#X_test = scaler.transform(X_test)
|
||||
|
||||
# run NN on train dat and evaluate on test data
|
||||
clf = MLPClassifier()
|
||||
clf.fit(X,y)
|
||||
clf.score(X_test,y_test)
|
||||
|
||||
# save NN model
|
||||
joblib.dump(clf, os.path.join(os.getcwd(), 'sand_classification', 'NN1.pkl'))
|
||||
Loading…
Reference in New Issue