def clip_array_with_vector(
array, array_affine, geometries, inverted=False, clip_buffer=0
):
"""
Clip input array with a vector list.
Parameters
----------
array : array
input raster data
array_affine : Affine
Affine object describing the raster's geolocation
geometries : iterable
iterable of dictionaries, where every entry has a 'geometry' and
'properties' key.
inverted : bool
invert clip (default: False)
clip_buffer : integer
buffer (in pixels) geometries before clipping
Returns
-------
clipped array : array
"""
# buffer input geometries and clean up
buffered_geometries = []
for feature in geometries:
feature_geom = to_shape(feature["geometry"])
if feature_geom.is_empty:
continue
if feature_geom.geom_type == "GeometryCollection":
# for GeometryCollections apply buffer to every subgeometry
# and make union
buffered_geom = unary_union([g.buffer(clip_buffer) for g in feature_geom])
else:
buffered_geom = feature_geom.buffer(clip_buffer)
if not buffered_geom.is_empty:
buffered_geometries.append(buffered_geom)
# mask raster by buffered geometries
if buffered_geometries:
if array.ndim == 2:
return ma.masked_array(
array, geometry_mask(
buffered_geometries, array.shape, array_affine,
invert=inverted))
elif array.ndim == 3:
mask = geometry_mask(
buffered_geometries,
(array.shape[1], array.shape[2]),
array_affine,
invert=inverted
)
return ma.masked_array(array, mask=np.stack([mask for band in array]))
# if no geometries, return unmasked array
else:
fill = False if inverted else True
return ma.masked_array(array, mask=np.full(array.shape, fill, dtype=bool))
def test_geometry_invalid_geom(geom):
"""An invalid geometry should fail"""
with pytest.raises(ValueError) as exc_info, pytest.warns(ShapeSkipWarning):
geometry_mask([geom],
out_shape=DEFAULT_SHAPE,
transform=Affine.identity())
assert 'No valid geometry objects found for rasterize' in exc_info.value.args[
0]
def test_geometry_mask_invalid_shape(basic_geometry):
"""A width==0 or height==0 should fail with ValueError"""
for shape in [(0, 0), (1, 0), (0, 1)]:
with pytest.raises(ValueError) as exc_info:
geometry_mask([basic_geometry],
out_shape=shape,
transform=Affine.identity())
assert 'must be > 0' in exc_info.value.args[0]
def test_geometry_mask_invalid_shape(basic_geometry):
"""A width==0 or height==0 should fail with ValueError"""
for shape in [(0, 0), (1, 0), (0, 1)]:
with pytest.raises(ValueError) as exc_info:
geometry_mask(
[basic_geometry],
out_shape=shape,
transform=Affine.identity())
assert 'must be > 0' in exc_info.value.args[0]
def read_raster_from_polygon(
src: rio.DatasetReader,
poly: Union[Polygon, MultiPolygon]) -> np.ma.MaskedArray:
"""Read valid pixel values from all locations inside a polygon
Uses the polygon as a mask in addition to the existing raster mask
Args:
src: an open rasterio dataset to read from
poly: a shapely Polygon or MultiPolygon
Returns:
masked array of shape (nbands, nrows, ncols)
"""
# get the read parameters
window = rio.windows.from_bounds(*poly.bounds, src.transform)
transform = rio.windows.transform(window, src.transform)
# get the data
data = src.read(window=window, masked=True, boundless=True)
bands, rows, cols = data.shape
poly_mask = geometry_mask([poly],
transform=transform,
out_shape=(rows, cols))
# update the mask
data[:, poly_mask] = np.ma.masked
return data
def _clip_xarray(xds, geometries, all_touched, drop, invert):
"""
clip the xarray DataArray
"""
clip_mask_arr = geometry_mask(
geometries=geometries,
out_shape=(int(xds.rio.height), int(xds.rio.width)),
transform=xds.rio.transform(recalc=True),
invert=not invert,
all_touched=all_touched,
)
clip_mask_xray = xarray.DataArray(
clip_mask_arr,
dims=(xds.rio.y_dim, xds.rio.x_dim),
)
cropped_ds = xds.where(clip_mask_xray)
if drop:
cropped_ds.rio.set_spatial_dims(x_dim=xds.rio.x_dim,
y_dim=xds.rio.y_dim,
inplace=True)
cropped_ds = cropped_ds.rio.isel_window(
rasterio.windows.get_data_window(
np.ma.masked_array(clip_mask_arr, ~clip_mask_arr)))
if xds.rio.nodata is not None and not np.isnan(xds.rio.nodata):
cropped_ds = cropped_ds.fillna(xds.rio.nodata)
return cropped_ds.astype(xds.dtype)
def mask_rast_from_geom(raster_file, geom_iterator):
"""Returns a numpy mask of the same dimensions as the input raster and also
a numpy array copy of the input raster.
requires inputs:
- (gdal) raster file
- iterator of json like geometries
returns:
- a (numpy) raster mask
- a numpy array copy of the input raster (should be handled elsewhere?)
"""
with rasterio.open(raster_file) as src_rst:
dsm = src_rst.read()
out_meta = src_rst.meta.copy()
print(out_meta)
out_meta.update(dtype=rasterio.float32, driver='GTiff')
mask = features.geometry_mask(
[feature["feature"]["geometry"] for feature in geom_iterator],
src_rst.shape,
transform=src_rst.transform,
all_touched=True,
invert=True)
new_dsm = np.copy(np.squeeze(
dsm)) #Forstaer ikke hvorfor, men dsm'en har en ekstra dimension,
#som jeg fjerner med squeeze, saa den passer med result dsm'en
return mask, new_dsm
def get_slope_raster(dc, x_range, y_range, inputcrs, resolution,
geom_selectedarea):
ds_dem = dc.load(product='dem',
x=x_range,
y=y_range,
crs=inputcrs,
output_crs=inputcrs,
resolution=resolution,
dask_chunks={'time': 1})
# Construct a mask to only select pixels within the drawn polygon
mask_dem = features.geometry_mask(
[geom_selectedarea for geoms in [geom_selectedarea]],
out_shape=ds_dem.geobox.shape,
transform=ds_dem.geobox.affine,
all_touched=False,
invert=True)
# Calculate Slope category
slope = get_slope(ds_dem.band1.squeeze('time', drop=True),
*resolution).where(mask_dem)
slope_cat = xr.apply_ufunc(slope_category,
slope,
vectorize=True,
dask='parallelized',
output_dtypes=[np.float32])
slope_cat.name = 'slope_category'
return slope_cat
def get_dlcd_raster(dc, x_range, y_range, inputcrs, resolution,
geom_selectedarea):
"""
:param x_range: tuple() of x values
:param y_range: tuple of y values
:param inputcrs: crs of x and y values
:param resolution:
:param geom_selectedarea:
:return:
"""
ds_dlcd = dc.load(product='dlcdnsw',
x=x_range,
y=y_range,
crs=inputcrs,
output_crs=inputcrs,
resolution=resolution,
dask_chunks={'time': 1})
mask_dlcd = features.geometry_mask(
[geom_selectedarea for geoms in [geom_selectedarea]],
out_shape=ds_dlcd.geobox.shape,
transform=ds_dlcd.geobox.affine,
all_touched=False,
invert=True)
masked_dlcd = ds_dlcd.band1.where(mask_dlcd)
masked_dlcd.name = 'dlcd'
return masked_dlcd
def test_geometry_mask_invert(basic_geometry, basic_image_2x2):
assert np.array_equal(
basic_image_2x2,
geometry_mask([basic_geometry],
out_shape=DEFAULT_SHAPE,
transform=Affine.identity(),
invert=True))
def test_geometry_mask(basic_geometry, basic_image_2x2):
with rasterio.Env():
assert np.array_equal(
basic_image_2x2 == 0,
geometry_mask([basic_geometry],
out_shape=DEFAULT_SHAPE,
transform=Affine.identity()))
def get_income_level_segmentation_mask(labels_dict, levels, image_shape,
image_transform):
"""
Generate a roof segmentation mask from a dictionary of geometries
:param labels_dict: labels dictionary
:type labels_dict: dict
:param levels: levels - mask index dictionary
:type levels: dict
:param image_shape: satellite image patch shape
:type image_shape: tuple
:param image_transform: satellite image patch affine transform
:type image_transform: any
"""
mask = np.ndarray(
(len(levels), image_shape[0], image_shape[1])).astype(int)
for level, labels in labels_dict.items():
mask[levels[level]] = geometry_mask(labels, image_shape, image_transform, all_touched=False, invert=True) \
if labels else np.zeros(image_shape)
return mask
def rasterize(feature_collection, transform, out_shape, name='mask'):
"""
Transform vector geometries to raster form, return band sample where
raster is np.array of bool dtype (`True` value correspond to objects area)
Args:
feature_collection: `FeatureCollection` object
transform: Affine transformation object
Transformation from pixel coordinates of `source` to the
coordinate system of the input `shapes`. See the `transform`
property of dataset objects.
out_shape: tuple or list
Shape of output numpy ndarray.
name: output sample name, default `mask`
Returns:
`BandSample` object
"""
if len(feature_collection) > 0:
geometries = (f.geometry for f in feature_collection)
mask = geometry_mask(geometries,
out_shape=out_shape,
transform=transform,
invert=True).astype('uint8')
else:
mask = np.zeros(out_shape, dtype='uint8')
return BandSample(name, mask, feature_collection.crs, transform)
def clip_with_shape(raster, shape):
if isinstance(raster,str):
raster = rasterio.open(raster)
window = raster.window(*shape.bounds)
out_transform = raster.window_transform(window)
out_image = raster.read(window=window, masked=True)
out_shape = out_image.shape[1:]
if out_shape[0] * out_shape[1] > 0:
shape_mask = geometry_mask([shape], transform=out_transform, invert=False,
out_shape=out_shape, all_touched=False)
out_image.mask = out_image.mask | shape_mask
out_image.fill_value = None
out_image = out_image.filled(np.nan)
#out_image, out_transform = mask.mask(src, [Polygon(geoms_inters[41]['geometry']['coordinates'][0])], crop=True)
out_meta = raster.meta.copy()
# GTiff
out_meta.update({"driver": "GTiff",
"height": out_image.shape[1],
"width": out_image.shape[2],
"transform": out_transform,
"count": 2,
'compression': 'lzw',
'tiled': True
})
return (out_image, out_transform, out_meta)
else:
return [False]
def extract_values(pycno_array, gdf, transform, fieldname='Estimate'):
"""Extract raster value sums according to a provided polygon geodataframe
Args:
pycno_array (numpy array): 2D numpy array of pycnophylactic surface.
gdf (geopandas.geodataframe.GeoDataFrame): Target GeoDataFrame.
transform (rasterio geotransform): Relevant transform from pycno()
fieldname (str, optional): New gdf field to save estimates in. Default name: 'Estimate'.
Returns:
geopandas.geodataframe.GeoDataFrame: Target GeoDataFrame with appended estimates.
"""
from numpy import nansum
from rasterio.features import geometry_mask
out = gdf.copy()
estimates = []
# Iterate through geodataframe and extract values
for idx, geom in gdf['geometry'].iteritems():
mask = geometry_mask([geom],
pycno_array.shape,
transform=transform,
invert=True)
estimates.append(nansum(pycno_array[mask]))
out[fieldname] = estimates
return out
def shapefile_mask(dataset: xr.Dataset, shapefile) -> np.array:
"""
Extracts a mask from a shapefile using dataset latitude and longitude extents.
Args:
dataset (xarray.Dataset): The dataset with latitude and longitude extents.
shapefile (string): The shapefile to be used.
Returns:
A boolean mask array.
"""
import pyproj
with fiona.open(shapefile, 'r') as src:
collection = list(src)
geometries = []
for feature in collection:
geom = shape(feature['geometry'])
project = partial(
pyproj.transform,
pyproj.Proj(init=src.crs['init']), # source crs
pyproj.Proj(init='epsg:4326')) # destination crs
geom = transform(project, geom) # apply projection
geometries.append(geom)
geobox = dataset.geobox
mask = geometry_mask(
geometries,
out_shape=geobox.shape,
transform=geobox.affine,
all_touched=True,
invert=True)
return mask
def get_roof_segmentation_mask(labels_dict, image_shape, image_transform):
"""
Generate a roof segmentation mask from a dictionary of geometries
:param labels_dict: labels dictionary
:type labels_dict: dict
:param image_shape: satellite image patch shape
:type image_shape: tuple
:param image_transform: satellite image patch affine transform
:type image_transform: any
"""
geometries = []
for _, labels in labels_dict.items():
geometries += labels
if geometries:
mask = geometry_mask(geometries,
image_shape,
image_transform,
all_touched=True,
invert=True)
else:
mask = np.zeros(image_shape).astype(int)
return mask
def mask_geom_on_raster(raster_data: np.ndarray, shifted_affine: Affine,
geom: BasePolygon) -> np.ndarray:
""""
For a given polygon, returns a numpy masked array with the intersecting
values of the raster at `raster_path` unmasked, all non-intersecting
cells are masked. This assumes that the input geometry is in the same
SRS as the raster. Currently only reads from a single band.
Args:
geom (Shapely Geometry): A polygon in the same SRS as `raster_path`
which will define the area of the raster to mask.
raster_path (string): A local file path to a geographic raster
containing values to extract.
Returns
Numpy masked array of source raster, cropped to the extent of the
input geometry, with any modifications applied. Areas where the
supplied geometry does not intersect are masked.
"""
if raster_data.size > 0:
# Create a numpy array to mask cells which don't intersect with the
# polygon. Cells that intersect will have value of 1 (unmasked), the
# rest are filled with 0s (masked)
geom_mask = features.geometry_mask([geom],
out_shape=raster_data.shape,
transform=shifted_affine,
invert=True)
# Mask the data array, with modifications applied, by the query polygon
return geom_mask
else:
return np.array([])
def test_geometry_invalid_geom():
"""An invalid geometry should fail"""
invalid_geoms = [
{'type': 'Invalid'}, # wrong type
{'type': 'Point'}, # missing coordinates
{'type': 'Point', 'coordinates': []} # empty coordinates
]
for geom in invalid_geoms:
with pytest.raises(ValueError) as exc_info:
geometry_mask(
[geom],
out_shape=DEFAULT_SHAPE,
transform=Affine.identity())
assert 'Invalid geometry' in exc_info.value.args[0]
def test_geometry_invalid_geom():
"""An invalid geometry should fail"""
invalid_geoms = [
{'type': 'Invalid'}, # wrong type
{'type': 'Point'}, # missing coordinates
{'type': 'Point', 'coordinates': []} # empty coordinates
]
for geom in invalid_geoms:
with pytest.raises(ValueError) as exc_info:
geometry_mask(
[geom],
out_shape=DEFAULT_SHAPE,
transform=Affine.identity())
assert 'Invalid geometry' in exc_info.value.args[0]
def test_geometry_mask():
rows = cols = 10
transform = (1.0, 0.0, 0.0, 0.0, -1.0, 0.0)
truth = numpy.zeros((rows, cols), dtype=rasterio.bool_)
truth[2:5, 2:5] = True
with rasterio.drivers():
s = shapes((truth * 10).astype(rasterio.ubyte), transform=transform)
# Strip out values returned from shapes, and only keep first shape
geoms = [next(s)[0]]
# Regular mask should be the inverse of truth raster
mask = geometry_mask(geoms, out_shape=(rows, cols), transform=transform)
assert numpy.array_equal(mask, numpy.invert(truth))
# Inverted mask should be the same as the truth raster
mask = geometry_mask(geoms, out_shape=(rows, cols), transform=transform,
invert=True)
assert numpy.array_equal(mask, truth)
def test_geometry_mask(basic_geometry, basic_image_2x2):
assert np.array_equal(
basic_image_2x2 == 0,
geometry_mask(
[basic_geometry],
out_shape=DEFAULT_SHAPE,
transform=Affine.identity()
)
)
def rasterize(geometries, transform, shape):
"""Convert geometries to raster mask"""
if len(geometries) > 0:
mask = geometry_mask(geometries,
out_shape=shape,
transform=transform,
invert=True).astype('uint8')
else:
mask = np.zeros(shape, dtype='uint8')
return mask
def test_geometry_mask(basic_geometry, basic_image_2x2):
with rasterio.drivers():
assert numpy.array_equal(
basic_image_2x2 == 0,
geometry_mask(
[basic_geometry],
out_shape=DEFAULT_SHAPE,
transform=Affine.identity()
)
)
def viewshed(vrt,list_of_dicts,distance,point,
observer_height,grassdb,burn_viewshed_rst,total_cells_output):
## deles op - foerste funktion
with rasterio.open(vrt) as src_rst:
dsm = src_rst.read()
out_meta = src_rst.meta.copy()
print(out_meta)
out_meta.update(dtype=rasterio.int16,driver='GTiff')
mask = features.geometry_mask(
[feature["feature"]["geometry"] for feature in
list_of_dicts],
src_rst.shape,
transform=src_rst.transform,
all_touched=True,
invert=True)
new_dsm = np.copy(np.squeeze(dsm)) #Forstaer ikke hvorfor, men dsm'en har en ekstra dimension,
#som jeg fjerner med squeeze, saa den passer med result dsm'en
## deles op - anden funktion, maaske
with rasterio.Env():
result = features.rasterize(
((feature['feature']['geometry'],np.int(feature['feature']['properties']['hoejde'])
* 1000)
for feature in list_of_dicts),
out_shape=src_rst.shape,
transform=src_rst.transform,
all_touched=True)
new_dsm[mask] = result[mask]
## deles op - tredje funktion
with Session(gisdb=grassdb, location="test",create_opts=vrt):
import grass.script.array as garray
r_viewshed = Module('r.viewshed')
r_out_gdal = Module('r.out.gdal')
r_stats = Module('r.stats')
r_univar = Module('r.univar')
from_np_raster = garray.array()
from_np_raster[...] = new_dsm
from_np_raster.write('ny_rast',overwrite=True)
print(from_np_raster)
gcore.run_command('r.viewshed', overwrite=True, memory=2000,
input='ny_rast', output='viewshed', max_distance=distance,
coordinates=point, observer_elevation=observer_height)
r_stats(flags='nc',overwrite=True,input='viewshed',output=total_cells_output)
## finde ud af hvordan r_stats kan outputte til noget som
## python kan laese direkte
with open(total_cells_output) as tcls:
counts = []
for line in tcls:
nbr = int(line.split()[-1])
counts.append(nbr)
# summary = r_univar(map='viewshed')
#r_viewshed(input=from_np_raster, output='viewshed', max_distance=1000, memory=1424, coordinates=(701495,6201503), observer_elevation=500.0)
r_out_gdal(overwrite=True, input='viewshed', output=burn_viewshed_rst)
return sum(counts) #visible_cells
def geometry_mask(geom, data, affine, all_touched=True):
# Create a numpy array to mask cells which don't intersect with the
# polygon. Cells that intersect will have value of 0 (unmasked), the
# rest are filled with 1s (masked)
geom_mask = features.geometry_mask([geom],
out_shape=data.shape,
transform=affine,
all_touched=all_touched)
# Mask the data array, with modifications applied, by the query polygon
return np.ma.array(data=data, mask=geom_mask)
def test_geometry_mask_invert(basic_geometry, basic_image_2x2):
with Env():
assert np.array_equal(
basic_image_2x2,
geometry_mask(
[basic_geometry],
out_shape=DEFAULT_SHAPE,
transform=Affine.identity(),
invert=True
)
)
def test_rasterize():
"""
Test roundtripping from image mask to polygon
and back again using rasterio's built-in functions.
"""
image_shape = (200, 100)
coords = [(33, 27), (50, 22), (80, 55)]
coords.append(coords[0])
poly = Polygon(shell=coords)
p = mapping(poly)
# Create initial mask
mask0 = geometry_mask((p, ), image_shape, Affine.identity(), invert=True)
m = mask0.astype(N.uint8)
s = [geom for geom, value in shapes(m) if value == 1]
mask1 = geometry_mask(s, image_shape, Affine.identity(), invert=True)
assert mask0.sum() == mask1.sum()
assert N.allclose(mask0, mask1)
def get_image_and_mask(self, tile_geometry, debug_base_file_name=None):
km2_to_m2 = 1000.0 * 1000.0
surface_area_m2 = tile_geometry.area * km2_to_m2
min_tile_e = int(tile_geometry.bounds[0])
min_tile_n = int(tile_geometry.bounds[1])
max_tile_e = int(tile_geometry.bounds[2])
max_tile_n = int(tile_geometry.bounds[3])
image_bgr = self.download_image(max_tile_e, max_tile_n, min_tile_e,
min_tile_n)
# [a, b, d, e, xoff, yoff]
# x' = a * x + b * y + xoff
# y' = d * x + e * y + yoff
m = [
self.__final_tile_size, 0, 0, self.__final_tile_size,
-min_tile_e * self.__final_tile_size,
-min_tile_n * self.__final_tile_size
]
affine_geometry = affine_transform(tile_geometry, m)
min_x = floor(affine_geometry.bounds[0])
min_y = floor(affine_geometry.bounds[1])
max_x = floor(affine_geometry.bounds[2])
max_y = floor(affine_geometry.bounds[3])
max_y_vertically_flipped = image_bgr.shape[0] - 1 - min_y
min_y_vertically_flipped = image_bgr.shape[0] - 1 - max_y
affine = rasterio.Affine(1, 0, min_x, 0, -1, max_y)
pixels_within_geometry = geometry_mask(
[affine_geometry],
(max_y_vertically_flipped - min_y_vertically_flipped + 1,
max_x - min_x + 1),
affine,
invert=True)
image_bgri_cropped = image_bgr[
min_y_vertically_flipped:max_y_vertically_flipped + 1,
min_x:max_x + 1, :]
tile_file_name = None
if debug_base_file_name is not None:
centre_point = tile_geometry.centroid
tile_code = tile_eastings_and_northings_to_tile_code(
centre_point.x, centre_point.y)
tile_file_name = debug_base_file_name + '-' + tile_code
return image_bgri_cropped, pixels_within_geometry, surface_area_m2, tile_file_name
def xarray_geomask(
ds: Union[xr.Dataset, xr.DataArray],
geometry: Union[Polygon, MultiPolygon, Tuple[float, float, float,
float]],
geo_crs: str,
ds_dims: Tuple[str, str] = ("y", "x"),
):
"""Mask a ``xarray.Dataset`` based on a geometry.
Parameters
----------
ds : xarray.Dataset or xarray.DataArray
The dataset(array) to be masked
geometry : Polygon, MultiPolygon, or tuple of length 4
The geometry or bounding box to mask the data
geo_crs : str
The spatial reference of the input geometry
ds_dims : tuple of str, optional
The names of the vertical and horizontal dimensions (in that order)
of the dataset, default to ("y", "x").
Returns
-------
xarray.Dataset or xarray.DataArray
The input dataset with a mask applied (np.nan)
"""
if not isinstance(ds_dims, tuple) or len(ds_dims) != 2:
raise InvalidInputType("ds_dims", "tuple of length 2", '("y", "x")')
if "crs" not in ds.attrs:
raise ValueError("The input dataset is missing the crs attribute.")
_geometry = geo2polygon(geometry, geo_crs, ds.crs)
height, width = ds.sizes[ds_dims[0]], ds.sizes[ds_dims[1]]
transform = get_transform(_geometry, height, width)
_mask = rio_features.geometry_mask([_geometry], (height, width),
transform,
invert=True)
coords = {
ds_dims[0]: ds.coords[ds_dims[0]],
ds_dims[1]: ds.coords[ds_dims[1]]
}
mask = xr.DataArray(_mask, coords, dims=ds_dims)
ds_masked = ds.where(mask, drop=True)
ds_masked.attrs["transform"] = transform
ds_masked.attrs["bounds"] = _geometry.bounds
return ds_masked
def get_image_and_mask(self, tile_geometry, debug_base_file_name=None):
surface_area = tile_geometry.area
min_tile_y = int(tile_geometry.bounds[0])
min_tile_x = int(tile_geometry.bounds[1])
max_tile_y = int(tile_geometry.bounds[2])
max_tile_x = int(tile_geometry.bounds[3])
image_bgr = self.download_image(max_tile_y, max_tile_x, min_tile_y,
min_tile_x)
# [a, b, d, e, xoff, yoff]
# x' = a * x + b * y + xoff
# y' = d * x + e * y + yoff
m = [
0, self.__tile_size, self.__tile_size, 0.0,
-min_tile_x * self.__tile_size, -min_tile_y * self.__tile_size
]
affine_geometry = affine_transform(tile_geometry, m)
min_y = floor(affine_geometry.bounds[1])
min_x = floor(affine_geometry.bounds[0])
max_y = floor(affine_geometry.bounds[3])
max_x = floor(affine_geometry.bounds[2])
affine = rasterio.Affine(1, 0, min_x, 0, 1, min_y)
pixels_within_geometry = geometry_mask(
[affine_geometry], (max_y - min_y + 1, max_x - min_x + 1),
affine,
invert=True)
image_bgr_cropped = image_bgr[min_y:max_y + 1, min_x:max_x + 1, :]
tile_file_name = None
if debug_base_file_name is not None:
centre_point = tile_geometry.centroid
centre_tile_x = int(centre_point.x)
centre_tile_y = int(centre_point.y)
centre_pixel_x = int(
(centre_point.x - centre_tile_x) * self.__tile_size)
centre_pixel_y = int(
(centre_point.y - centre_tile_y) * self.__tile_size)
tile_code = f'{centre_tile_x}_{centre_pixel_x}={centre_tile_y}_{centre_pixel_y}'
tile_file_name = debug_base_file_name + '-' + tile_code
return image_bgr_cropped, pixels_within_geometry, surface_area, tile_file_name
def clip_array_with_vector(
array,
array_affine,
geometries,
inverted=False,
clip_buffer=0
):
"""
Clips input array with a vector list.
"""
buffered_geometries = []
for feature in geometries:
geom = shape(feature['geometry']).buffer(clip_buffer)
if not isinstance(geom, (Polygon, MultiPolygon, GeometryCollection)):
break
if geom.is_empty:
break
if isinstance(geom, GeometryCollection):
polygons = [
subgeom
for subgeom in geom
if isinstance(subgeom, (Polygon, MultiPolygon))
]
if not polygons:
break
new_geom = MultiPolygon(polygons)
geom = new_geom
buffered_geometries.append(geom)
if buffered_geometries:
mask = geometry_mask(
buffered_geometries,
array.shape,
array_affine,
invert=inverted
)
else:
if inverted:
fill = False
else:
fill = True
mask = np.full(array.shape, fill, dtype=bool)
return ma.masked_array(array, mask)
def extract_area(geom, dem, **kwargs):
# RasterIO's image-reading algorithm uses the location
# of polygon centers to determine the extent of polygons
msk = geometry_mask(
(mapping(geom),),
dem.shape,
dem.transform,
invert=True)
# shrink mask to the minimal area for efficient extraction
offset, msk = offset_mask(msk)
window = tuple((o,o+s)
for o,s in zip(offset,msk.shape))
# Currently just for a single band
# We could generalize to multiple
# bands if desired
z = dem.read(1,
window=window,
masked=True)
# mask out unused area
z[msk == False] = N.ma.masked
# Make vectors of rows and columns
rows, cols = (N.arange(first,last,1)
for first,last in window)
# 2d arrays of x,y,z
z = z.flatten()
xyz = [i.flatten()
for i in N.meshgrid(cols,rows)] + [z]
x,y,z = tuple(i[z.mask == False] for i in xyz)
# Transform into 3xn matrix of
# flattened coordinate values
coords = N.vstack((x,y,N.ones(z.shape)))
# Get affine transform for pixel centers
affine = dem.transform * Affine.translation(0.5, 0.5)
# Transform coordinates to DEM's reference
_ = N.array(affine).reshape((3,3))
coords = N.dot(_,coords)
coords[2] = z
return coords.transpose()
def extract_area(geom, dem, **kwargs):
# RasterIO's image-reading algorithm uses the location
# of polygon centers to determine the extent of polygons
msk = geometry_mask(
(mapping(geom),),
dem.shape,
dem.transform,
invert=True)
# shrink mask to the minimal area for efficient extraction
offset, msk = offset_mask(msk)
window = tuple((o,o+s)
for o,s in zip(offset,msk.shape))
# Currently just for a single band
# We could generalize to multiple
# bands if desired
z = dem.read(1,
window=window,
masked=True)
# mask out unused area
z[msk == False] = N.ma.masked
# Make vectors of rows and columns
rows, cols = (N.arange(first,last,1)
for first,last in window)
# 2d arrays of x,y,z
z = z.flatten()
xyz = [i.flatten()
for i in N.meshgrid(cols,rows)] + [z]
x,y,z = tuple(i[z.mask == False] for i in xyz)
# Transform into 3xn matrix of
# flattened coordinate values
coords = N.vstack((x,y,N.ones(z.shape)))
# Get affine transform for pixel centers
affine = dem.transform * Affine.translation(0.5, 0.5)
# Transform coordinates to DEM's reference
_ = N.array(affine).reshape((3,3))
coords = N.dot(_,coords)
coords[2] = z
return coords.transpose()
def determine_mosaic_quads_for_geometry(geometry: dict) -> List[str]:
# Parameters
width = MOSAIC_TILE_SIZE * 2 * abs(WEBM_ORIGIN) / (2**MOSAIC_LEVEL * 256)
num_tiles = int(2.0**MOSAIC_LEVEL * 256 / MOSAIC_TILE_SIZE)
# Generate a grid where values are True if the geometry is present and False otherwise
transform = rio.transform.from_origin(WEBM_ORIGIN, -WEBM_ORIGIN, width,
width)
shape = shapely.geometry.shape(geometry)
grid = np.flipud(
geometry_mask([shape], (num_tiles, num_tiles),
transform,
all_touched=True,
invert=True))
# Get quad labels
quads = list()
norths, easts = np.where(grid)
for north, east in zip(norths, easts):
quads.append('L15-{:04d}E-{:04d}N'.format(east, north))
return quads
def preprocessed_hazardlevel(self, geometry):
hazardlevel = None
reader = self.readers[0]
for polygon in geometry.geoms:
if not polygon.intersects(self.bbox):
continue
window = reader.window(*polygon.bounds)
data = reader.read(1, window=window, masked=True)
if data.shape[0] * data.shape[1] == 0:
continue
if data.mask.all():
continue
geometry_mask = features.geometry_mask(
[polygon],
out_shape=data.shape,
transform=reader.window_transform(window),
all_touched=True,
)
data.mask = data.mask | geometry_mask
del geometry_mask
if data.mask.all():
continue
for level in ("HIG", "MED", "LOW", "VLO"):
level_obj = HazardLevel.get(self.dbsession, level)
if level_obj <= hazardlevel:
break
if level in self.type_settings["values"]:
values = self.type_settings["values"][level]
for value in values:
if value in data:
hazardlevel = level_obj
break
return hazardlevel
def get_segment_masks(image, polys, invert=True):
"""Image masks of the segments. Converted the segment polygons to binary images.
:param image: CatalogImage
:param polys: list of segment Shapely Polygons
:param invert: specify whether or not to invert the image mask
:returns: a list of image aois given the bounds of the segment and list of segment masks
"""
image_aoi_segs = []
seg_masks = []
for poly in polys:
bounds = poly.bounds
image_aoi_seg = image.aoi(bbox=list(bounds))
segment_mask = geometry_mask([poly], transform=image_aoi_seg.affine,
out_shape=(image_aoi_seg.shape[1], image_aoi_seg.shape[2]), invert=invert)
image_aoi_segs.append(image_aoi_seg)
seg_masks.append(segment_mask)
return image_aoi_segs, seg_masks
def preprocessed_hazardlevel(type_settings, layer, reader, geometry):
hazardlevel = None
for polygon in geometry.geoms:
window = reader.window(*polygon.bounds)
data = reader.read(1, window=window, masked=True)
if data.shape[0] * data.shape[1] == 0:
continue
if data.mask.all():
continue
geometry_mask = features.geometry_mask(
[polygon],
out_shape=data.shape,
transform=reader.window_transform(window),
all_touched=True)
data.mask = data.mask | geometry_mask
del geometry_mask
if data.mask.all():
continue
for level in (u'HIG', u'MED', u'LOW', u'VLO'):
level_obj = HazardLevel.get(level)
if level_obj <= hazardlevel:
break
if level in type_settings['values']:
values = type_settings['values'][level]
for value in values:
if value in data:
hazardlevel = level_obj
break
return hazardlevel
def mask(raster, shapes, nodata=None, crop=False, all_touched=False,
invert=False, pad=False):
"""Mask the area outside of the input shapes with nodata.
For all regions in the input raster outside of the regions defined by
`shapes`, sets any data present to nodata.
Parameters
----------
raster: rasterio RasterReader object
Raster to which the mask will be applied.
shapes: list of polygons
Polygons are GeoJSON-like dicts specifying the boundaries of features
in the raster to be kept. All data outside of specified polygons
will be set to nodata.
nodata: int or float (opt)
Value representing nodata within each raster band. If not set,
defaults to the nodata value for the input raster. If there is no
set nodata value for the raster, it defaults to 0.
crop: bool (opt)
Whether to crop the raster to the extent of the data. Defaults to
False.
all_touched: bool (opt)
Use all pixels touched by features. If False (default), use only
pixels whose center is within the polygon or that are selected by
Bresenhams line algorithm.
invert: bool (opt)
If True, mask will be True for pixels that overlap shapes.
False by default.
pad: bool (opt)
If True, the cropped output will be padded in each direction by
one half of a pixel. Defaults to False.
Returns
-------
tuple
Two elements:
masked : numpy ndarray
Data contained in raster after applying the mask.
out_transform : affine.Affine()
Information for mapping pixel coordinates in `masked` to another
coordinate system.
"""
if crop and invert:
raise ValueError("crop and invert cannot both be True.")
if nodata is None:
if raster.nodata is not None:
nodata = raster.nodata
else:
nodata = 0
# "North down" georeferencing or ungeoreferenced rasters require
# bounds shuffling.
north_up = raster.transform.e <= 0
# Calculate the bounds of all features.
all_bounds = [
rasterio.features.bounds(shape, north_up=north_up) for shape in shapes]
lefts, bottoms, rights, tops = zip(*all_bounds)
if pad:
dx = raster.res[0] / 2
dy = raster.res[1] / 2
else:
dx = 0.0
dy = 0.0
if north_up:
mask_bounds = (min(lefts) - dx, min(bottoms) - dy,
max(rights) + dx, max(tops) + dy)
else:
mask_bounds = (min(lefts) - dx, max(bottoms) + dy,
max(rights) + dx, min(tops) - dy)
source_bounds = raster.bounds
# Raise or warn about bounds mismatches.
if rasterio.coords.disjoint_bounds(source_bounds, mask_bounds):
if crop:
raise ValueError("Input shapes do not overlap raster.")
else:
warnings.warn("GeoJSON outside bounds of existing output " +
"raster. Are they in different coordinate " +
"reference systems?")
if crop:
bounds_window = raster.window(*mask_bounds)
# Get the window with integer height
# and width that contains the bounds window.
out_window = bounds_window.round_shape(op='ceil')
height = int(out_window.height)
width = int(out_window.width)
out_shape = (raster.count, height, width)
out_transform = raster.window_transform(out_window)
logger.debug("Out window: %r", out_window)
logger.debug("Out transform: %r", out_transform)
else:
out_window = None
out_shape = (raster.count, raster.height, raster.width)
out_transform = raster.transform
# Read the window of imagery.
out_image = raster.read(window=out_window, out_shape=out_shape,
masked=True)
mask_shape = out_image.shape[1:]
shape_mask = geometry_mask(shapes, transform=out_transform, invert=invert,
out_shape=mask_shape, all_touched=all_touched)
out_image.mask = out_image.mask | shape_mask
out_image.fill_value = nodata
for i in range(raster.count):
out_image[i] = out_image[i].filled(nodata)
return out_image, out_transform
def notpreprocessed_hazardlevel(hazardtype, type_settings, layers, readers,
geometry):
level_vlo = HazardLevel.get(u'VLO')
hazardlevel = None
# Create some optimization caches
polygons = {}
bboxes = {}
geometry_masks = {} # Storage for the geometry geometry_masks
inverted_comparison = ('inverted_comparison' in type_settings and
type_settings['inverted_comparison'])
for level in (u'HIG', u'MED', u'LOW'):
layer = layers[level]
reader = readers[level]
threshold = get_threshold(hazardtype,
layer.local,
layer.hazardlevel.mnemonic,
layer.hazardunit)
if threshold is None:
raise ProcessException(
'No threshold found for {} {} {} {}'
.format(hazardtype,
'local' if layer.local else 'global',
layer.hazardlevel.mnemonic,
layer.hazardunit))
for i in xrange(0, len(geometry.geoms)):
if i not in polygons:
polygon = geometry.geoms[i]
bbox = polygon.bounds
polygons[i] = polygon
bboxes[i] = bbox
else:
polygon = polygons[i]
bbox = bboxes[i]
window = reader.window(*bbox)
# data: MaskedArray
data = reader.read(1, window=window, masked=True)
# check if data is empty (cols x rows)
if data.shape[0] * data.shape[1] == 0:
continue
# all data is masked which means that all is NODATA
if data.mask.all():
continue
if inverted_comparison:
data = data < threshold
else:
data = data > threshold
# some hazard types have a specific mask layer with very low
# return period which should be used as mask for other layers
# for example River Flood
if ('mask_return_period' in type_settings):
mask = readers['mask'].read(1, window=window, masked=True)
if inverted_comparison:
mask = mask < threshold
else:
mask = mask > threshold
# apply the specific layer mask
data.mask = ma.getmaskarray(data) | mask.filled(False)
del mask
if data.mask.all():
continue
if i in geometry_masks:
geometry_mask = geometry_masks[i]
else:
geometry_mask = features.geometry_mask(
[polygon],
out_shape=data.shape,
transform=reader.window_transform(window),
all_touched=True)
geometry_masks[i] = geometry_mask
data.mask = ma.getmaskarray(data) | geometry_mask
del geometry_mask
# If at least one value is True this means that there's at least
# one raw value > threshold
if data.any():
hazardlevel = layer.hazardlevel
break
# check one last time is array is filled with NODATA
if data.mask.all():
continue
# Here we have at least one value lower than the current level
# threshold
if hazardlevel is None:
hazardlevel = level_vlo
# we got a value for the level, no need to go further, this will be
# the highest one
if hazardlevel == layer.hazardlevel:
break
return hazardlevel
def test_geometry_mask_no_transform(basic_geometry):
with pytest.raises(TypeError):
geometry_mask(
[basic_geometry],
out_shape=DEFAULT_SHAPE,
transform=None)
def notpreprocessed_hazardlevel(hazardtype, type_settings, layers, readers,
geometry):
level_VLO = HazardLevel.get(u'VLO')
hazardlevel = None
# Create some optimization caches
polygons = {}
bboxes = {}
masks = {}
inverted_comparison = ('inverted_comparison' in type_settings and
type_settings['inverted_comparison'])
for level in (u'HIG', u'MED', u'LOW'):
layer = layers[level]
reader = readers[level]
threshold = get_threshold(hazardtype,
layer.local,
layer.hazardlevel.mnemonic,
layer.hazardunit)
if threshold is None:
raise ProcessException(
'No threshold found for {} {} {} {}'
.format(hazardtype,
'local' if layer.local else 'global',
layer.hazardlevel.mnemonic,
layer.hazardunit))
for i in xrange(0, len(geometry.geoms)):
if i not in polygons:
polygon = geometry.geoms[i]
bbox = polygon.bounds
polygons[i] = polygon
bboxes[i] = bbox
else:
polygon = polygons[i]
bbox = bboxes[i]
window = reader.window(*bbox)
data = reader.read(1, window=window, masked=True)
if data.shape[0] * data.shape[1] == 0:
continue
if data.mask.all():
continue
if inverted_comparison:
data = data < threshold
else:
data = data > threshold
if ('mask_return_period' in type_settings):
mask = readers['mask'].read(1, window=window, masked=True)
if inverted_comparison:
mask = mask < threshold
else:
mask = mask > threshold
data.mask = ma.getmaskarray(data) | mask.filled(False)
if data.mask.all():
continue
if i in masks:
mask = masks[i]
else:
mask = features.geometry_mask(
[polygon],
out_shape=data.shape,
transform=reader.window_transform(window),
all_touched=True)
masks[i] = mask
data.mask = ma.getmaskarray(data) | mask
if data.any():
hazardlevel = layer.hazardlevel
break # No need to go further
if data.mask.all():
continue
if hazardlevel is None:
hazardlevel = level_VLO
if hazardlevel == layer.hazardlevel:
break # No need to go further
return hazardlevel
def mask(raster, shapes, nodata=None, crop=False, all_touched=False,
invert=False):
"""Mask the area outside of the input shapes with nodata.
For all regions in the input raster outside of the regions defined by
`shapes`, sets any data present to nodata.
Parameters
----------
raster: rasterio RasterReader object
Raster to which the mask will be applied.
shapes: list of polygons
Polygons are GeoJSON-like dicts specifying the boundaries of features
in the raster to be kept. All data outside of specified polygons
will be set to nodata.
nodata: int or float (opt)
Value representing nodata within each raster band. If not set,
defaults to the nodata value for the input raster. If there is no
set nodata value for the raster, it defaults to 0.
crop: bool (opt)
Whether to crop the raster to the extent of the data. Defaults to
False.
all_touched: bool (opt)
Use all pixels touched by features. If False (default), use only
pixels whose center is within the polygon or that are selected by
Bresenhams line algorithm.
invert: bool (opt)
If True, mask will be True for pixels that overlap shapes.
False by default.
Returns
-------
masked: numpy ndarray
Data contained in raster after applying the mask.
out_transform: affine object
Information for mapping pixel coordinates in `masked` to another
coordinate system.
"""
if crop and invert:
raise ValueError("crop and invert cannot both be True.")
if nodata is None:
if raster.nodata is not None:
nodata = raster.nodata
else:
nodata = 0
all_bounds = [rasterio.features.bounds(shape) for shape in shapes]
minxs, minys, maxxs, maxys = zip(*all_bounds)
mask_bounds = (min(minxs), min(minys), max(maxxs), max(maxys))
invert_y = raster.affine.e > 0
source_bounds = raster.bounds
if invert_y:
source_bounds = [source_bounds[0], source_bounds[3],
source_bounds[2], source_bounds[1]]
if rasterio.coords.disjoint_bounds(source_bounds, mask_bounds):
if crop:
raise ValueError("Input shapes do not overlap raster.")
else:
warnings.warn("GeoJSON outside bounds of existing output " +
"raster. Are they in different coordinate " +
"reference systems?")
if invert_y:
mask_bounds = [mask_bounds[0], mask_bounds[3],
mask_bounds[2], mask_bounds[1]]
if crop:
window = raster.window(*mask_bounds)
out_transform = raster.window_transform(window)
else:
window = None
out_transform = raster.affine
out_image = raster.read(window=window, masked=True)
out_shape = out_image.shape[1:]
shape_mask = geometry_mask(shapes, transform=out_transform, invert=invert,
out_shape=out_shape, all_touched=all_touched)
out_image.mask = out_image.mask | shape_mask
out_image.fill_value = nodata
for i in range(raster.count):
out_image[i] = out_image[i].filled(nodata)
return out_image, out_transform
def raster_geometry_mask(dataset, shapes, all_touched=False, invert=False,
crop=False, pad=False):
"""Create a mask from shapes, transform, and optional window within original
raster.
By default, mask is intended for use as a numpy mask, where pixels that
overlap shapes are False.
If shapes do not overlap the raster and crop=True, a ValueError is
raised. Otherwise, a warning is raised, and a completely True mask
is returned (if invert is False).
Parameters
----------
dataset: a dataset object opened in 'r' mode
Raster for which the mask will be created.
shapes: list of polygons
GeoJSON-like dict representation of polygons that will be used to
create the mask.
all_touched: bool (opt)
Include a pixel in the mask if it touches any of the shapes.
If False (default), include a pixel only if its center is within one of
the shapes, or if it is selected by Bresenham's line algorithm.
invert: bool (opt)
If False (default), mask will be `False` inside shapes and `True`
outside. If True, mask will be `True` inside shapes and `False`
outside.
crop: bool (opt)
Whether to crop the dataset to the extent of the shapes. Defaults to
False.
pad: bool (opt)
If True, the features will be padded in each direction by
one half of a pixel prior to cropping dataset. Defaults to False.
Returns
-------
tuple
Three elements:
mask : numpy ndarray of type 'bool'
Mask that is `True` outside shapes, and `False` within shapes.
out_transform : affine.Affine()
Information for mapping pixel coordinates in `masked` to another
coordinate system.
window: rasterio.windows.Window instance
Window within original raster covered by shapes. None if crop
is False.
"""
if crop and invert:
raise ValueError("crop and invert cannot both be True.")
if crop and pad:
pad_x = 0.5 # pad by 1/2 of pixel size
pad_y = 0.5
else:
pad_x = 0
pad_y = 0
north_up = dataset.transform.e <= 0
rotated = dataset.transform.b != 0 or dataset.transform.d != 0
try:
window = geometry_window(dataset, shapes, north_up=north_up, rotated=rotated,
pad_x=pad_x, pad_y=pad_y)
except WindowError:
# If shapes do not overlap raster, raise Exception or UserWarning
# depending on value of crop
if crop:
raise ValueError('Input shapes do not overlap raster.')
else:
warnings.warn('shapes are outside bounds of raster. '
'Are they in different coordinate reference systems?')
# Return an entirely True mask (if invert is False)
mask = np.ones(shape=dataset.shape[-2:], dtype='bool') * (not invert)
return mask, dataset.transform, None
if crop:
transform = dataset.window_transform(window)
out_shape = (int(window.height), int(window.width))
else:
window = None
transform = dataset.transform
out_shape = (int(dataset.height), int(dataset.width))
mask = geometry_mask(shapes, transform=transform, invert=invert,
out_shape=out_shape, all_touched=all_touched)
return mask, transform, window
def mask(
ctx,
files,
output,
geojson_mask,
driver,
all_touched,
crop,
invert,
creation_options):
"""Masks in raster using GeoJSON features (masks out all areas not covered
by features), and optionally crops the output raster to the extent of the
features. Features are assumed to be in the same coordinate reference
system as the input raster.
GeoJSON must be the first input file or provided from stdin:
> rio mask input.tif output.tif --geojson-mask features.json
> rio mask input.tif output.tif --geojson-mask - < features.json
If the output raster exists, it will be completely overwritten with the
results of this operation.
The result is always equal to or within the bounds of the input raster.
--crop and --invert options are mutually exclusive.
--crop option is not valid if features are completely outside extent of
input raster.
"""
from rasterio.features import geometry_mask
from rasterio.features import bounds as calculate_bounds
verbosity = (ctx.obj and ctx.obj.get('verbosity')) or 1
output, files = resolve_inout(files=files, output=output)
input = files[0]
if geojson_mask is None:
click.echo('No GeoJSON provided, INPUT will be copied to OUTPUT',
err=True)
shutil.copy(input, output)
return
if crop and invert:
click.echo('Invert option ignored when using --crop', err=True)
invert = False
with rasterio.drivers(CPL_DEBUG=verbosity > 2):
try:
with click.open_file(geojson_mask) as f:
geojson = json.loads(f.read())
except ValueError:
raise click.BadParameter('GeoJSON could not be read from '
'--geojson-mask or stdin',
param_hint='--geojson-mask')
if 'features' in geojson:
geometries = (f['geometry'] for f in geojson['features'])
elif 'geometry' in geojson:
geometries = (geojson['geometry'], )
else:
raise click.BadParameter('Invalid GeoJSON', param=input,
param_hint='input')
bounds = geojson.get('bbox', calculate_bounds(geojson))
with rasterio.open(input) as src:
# If y pixel value is positive, then invert y dimension in bounds
invert_y = src.affine.e > 0
src_bounds = src.bounds
if invert_y:
src_bounds = [src.bounds[0], src.bounds[3],
src.bounds[2], src.bounds[1]]
has_disjoint_bounds = disjoint_bounds(bounds, src_bounds)
if crop:
if has_disjoint_bounds:
raise click.BadParameter('not allowed for GeoJSON outside '
'the extent of the input raster',
param=crop, param_hint='--crop')
if invert_y:
bounds = (bounds[0], bounds[3], bounds[2], bounds[1])
window = src.window(*bounds)
transform = src.window_transform(window)
(r1, r2), (c1, c2) = window
mask_shape = (r2 - r1, c2 - c1)
else:
if has_disjoint_bounds:
click.echo('GeoJSON outside bounds of existing output '
'raster. Are they in different coordinate '
'reference systems?',
err=True)
window = None
transform = src.affine
mask_shape = src.shape
mask = geometry_mask(
geometries,
out_shape=mask_shape,
transform=transform,
all_touched=all_touched,
invert=invert)
meta = src.meta.copy()
meta.update(**creation_options)
meta.update({
'driver': driver,
'height': mask.shape[0],
'width': mask.shape[1],
'transform': transform
})
with rasterio.open(output, 'w', **meta) as out:
for bidx in range(1, src.count + 1):
img = src.read(bidx, masked=True, window=window)
img.mask = img.mask | mask
out.write_band(bidx, img.filled(src.nodatavals[bidx-1]))
