baseclasses.py 82.6 KB
Newer Older
1 2
# -*- coding: utf-8 -*-

3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
# geoarray, A fast Python interface for image geodata - either on disk or in memory.
#
# Copyright (C) 2019  Daniel Scheffler (GFZ Potsdam, daniel.scheffler@gfz-potsdam.de)
#
# This software was developed within the context of the GeoMultiSens project funded
# by the German Federal Ministry of Education and Research
# (project grant code: 01 IS 14 010 A-C).
#
# This program is free software: you can redistribute it and/or modify it under
# the terms of the GNU Lesser General Public License as published by the Free
# Software Foundation, either version 3 of the License, or (at your option) any
# later version.
#
# This program is distributed in the hope that it will be useful, but WITHOUT
# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
# FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more
# details.
#
# You should have received a copy of the GNU Lesser General Public License along
# with this program.  If not, see <http://www.gnu.org/licenses/>.


25 26
import os
import warnings
27
from pkgutil import find_loader
28
from collections import OrderedDict
Daniel Scheffler's avatar
Daniel Scheffler committed
29
from copy import deepcopy
30
from typing import Union  # noqa F401
31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48

import numpy as np
from matplotlib import pyplot as plt
from osgeo import gdal_array
# custom
from shapely.geometry import Polygon
from shapely.wkt import loads as shply_loads
from six import PY3

# mpl_toolkits.basemap -> imported when GeoArray.show_map() is used
# dill -> imported when dumping GeoArray

try:
    from osgeo import gdal
    from osgeo import gdalnumeric
except ImportError:
    import gdal
    import gdalnumeric
49 50 51 52
from py_tools_ds.convenience.object_oriented import alias_property
from py_tools_ds.geo.coord_calc import get_corner_coordinates
from py_tools_ds.geo.coord_grid import snap_bounds_to_pixGrid
from py_tools_ds.geo.coord_trafo import mapXY2imXY, imXY2mapXY, transform_any_prj, reproject_shapelyGeometry
53
from py_tools_ds.geo.projection import prj_equal, WKT2EPSG, EPSG2WKT, isLocal
54 55
from py_tools_ds.geo.raster.conversion import raster2polygon
from py_tools_ds.geo.vector.topology \
56
    import get_footprint_polygon, polyVertices_outside_poly, fill_holes_within_poly
57 58
from py_tools_ds.geo.vector.geometry import boxObj
from py_tools_ds.io.raster.gdal import get_GDAL_ds_inmem
59
from py_tools_ds.compatibility.gdal import get_gdal_func
60
from py_tools_ds.numeric.numbers import is_number
61
from py_tools_ds.numeric.array import get_array_tilebounds
62 63 64

#  internal imports
from .subsetting import get_array_at_mapPos
65
from .metadata import GDAL_Metadata
66

67
if PY3:
68
    # noinspection PyCompatibility
69 70 71
    from builtins import TimeoutError, FileNotFoundError
else:
    from py_tools_ds.compatibility.python.exceptions import TimeoutError, FileNotFoundError
72

73
__author__ = 'Daniel Scheffler'
74 75 76


class GeoArray(object):
Daniel Scheffler's avatar
Daniel Scheffler committed
77 78 79 80 81 82 83
    """
    This class creates a fast Python interface for geodata - either on disk or in memory. It can be instanced
    with a file path or with a numpy array and the corresponding geoinformation. Instances can always be indexed
    like normal numpy arrays, no matter if GeoArray has been instanced from file or from an in-memory array.
    GeoArray provides a wide range of geo-related attributes belonging to the dataset as well as some functions for
    quickly visualizing the data as a map, a simple image or an interactive image.
    """
84 85
    def __init__(self, path_or_array, geotransform=None, projection=None, bandnames=None, nodata=None, progress=True,
                 q=False):
86
        # type: (Union[str, np.ndarray, GeoArray], tuple, str, list, float, bool, bool) -> None
Daniel Scheffler's avatar
Daniel Scheffler committed
87
        """Get an instance of GeoArray.
88 89 90 91 92 93 94 95 96 97 98 99 100 101

        :param path_or_array:   a numpy.ndarray or a valid file path
        :param geotransform:    GDAL geotransform of the given array or file on disk
        :param projection:      projection of the given array or file on disk as WKT string
                                (only needed if GeoArray is instanced with an array)
        :param bandnames:       names of the bands within the input array, e.g. ['mask_1bit', 'mask_clouds'],
                                (default: ['B1', 'B2', 'B3', ...])
        :param nodata:          nodata value
        :param progress:        show progress bars (default: True)
        :param q:               quiet mode (default: False)
        """

        # TODO implement compatibility to GDAL VRTs
        if not (isinstance(path_or_array, (str, np.ndarray, GeoArray)) or
102
           issubclass(getattr(path_or_array, '__class__'), GeoArray)):
103
            raise ValueError("%s parameter 'arg' takes only string, np.ndarray or GeoArray(and subclass) instances. "
104
                             "Got %s." % (self.__class__.__name__, type(path_or_array)))
105 106

        if path_or_array is None:
107
            raise ValueError("The %s parameter 'path_or_array' must not be None!" % self.__class__.__name__)
108 109 110 111 112

        if isinstance(path_or_array, str):
            assert ' ' not in path_or_array, "The given path contains whitespaces. This is not supported by GDAL."

            if not os.path.exists(path_or_array):
113
                raise FileNotFoundError(path_or_array)
114

115 116
        if isinstance(path_or_array, GeoArray) or issubclass(getattr(path_or_array, '__class__'), GeoArray):
            self.__dict__ = path_or_array.__dict__.copy()
117
            self._initParams = dict([x for x in locals().items() if x[0] != "self"])
118 119
            self.geotransform = geotransform or self.geotransform
            self.projection = projection or self.projection
120
            self.bandnames = bandnames or list(self.bandnames.keys())
121 122 123
            self._nodata = nodata if nodata is not None else self._nodata
            self.progress = False if progress is False else self.progress
            self.q = q if q is not None else self.q
124 125

        else:
126 127 128 129 130 131 132 133 134 135 136 137 138 139 140
            self._initParams = dict([x for x in locals().items() if x[0] != "self"])
            self.arg = path_or_array
            self._arr = path_or_array if isinstance(path_or_array, np.ndarray) else None
            self.filePath = path_or_array if isinstance(path_or_array, str) and path_or_array else None
            self.basename = os.path.splitext(os.path.basename(self.filePath))[0] if not self.is_inmem else 'IN_MEM'
            self.progress = progress
            self.q = q
            self._arr_cache = None  # dict containing key 'pos' and 'arr_cached'
            self._geotransform = None
            self._projection = None
            self._shape = None
            self._dtype = None
            self._nodata = nodata
            self._mask_nodata = None
            self._mask_baddata = None
141 142
            self._footprint_poly = None
            self._gdalDataset_meta_already_set = False
143 144
            self._metadata = None
            self._bandnames = None
145 146

            if bandnames:
147
                self.bandnames = bandnames  # use property in order to validate given value
148
            if geotransform:
149
                self.geotransform = geotransform  # use property in order to validate given value
150
            if projection:
151
                self.projection = projection  # use property in order to validate given value
152 153 154 155 156 157 158 159 160 161

            if self.filePath:
                self.set_gdalDataset_meta()

    @property
    def arr(self):
        return self._arr

    @arr.setter
    def arr(self, ndarray):
162 163
        assert isinstance(ndarray, np.ndarray), "'arr' can only be set to a numpy array! Got %s." % type(ndarray)
        # assert ndarray.shape == self.shape, "'arr' can only be set to a numpy array with shape %s. Received %s. " \
164 165 166 167 168
        #                                    "If you need to change the dimensions, create a new instance of %s." \
        #                                    %(self.shape, ndarray.shape, self.__class__.__name__)
        #  THIS would avoid warping like this: geoArr.arr, geoArr.gt, geoArr.prj = warp(...)

        if ndarray.shape != self.shape:
169
            self.flush_cache()  # the cached array is not useful anymore
170 171 172 173 174

        self._arr = ndarray

    @property
    def bandnames(self):
175
        if self._bandnames and len(self._bandnames) == self.bands:
176 177
            return self._bandnames
        else:
178
            del self.bandnames  # runs deleter which sets it to default values
179 180 181 182 183 184 185
            return self._bandnames

    @bandnames.setter
    def bandnames(self, list_bandnames):
        # type: (list) -> None

        if list_bandnames:
186 187 188 189 190 191 192 193
            if not isinstance(list_bandnames, list):
                raise TypeError("A list must be given when setting the 'bandnames' attribute. "
                                "Received %s." % type(list_bandnames))
            if len(list_bandnames) != self.bands:
                raise ValueError('Number of given bandnames does not match number of bands in array.')
            if len(list(set([type(b) for b in list_bandnames]))) != 1 or not isinstance(list_bandnames[0], str):
                raise ValueError("'bandnames must be a set of strings. Got other datatypes in there.'")

194
            bN_dict = OrderedDict((band, i) for i, band in enumerate(list_bandnames))
195 196 197

            if len(bN_dict) != self.bands:
                raise ValueError('Bands must different names. Received band list: %s' % list_bandnames)
198 199 200

            self._bandnames = bN_dict

201
            try:
202
                self.metadata.band_meta['band_names'] = list_bandnames
203 204 205
            except AttributeError:
                # in case self._metadata is None
                pass
206 207 208 209 210 211 212
        else:
            del self.bandnames

    @bandnames.deleter
    def bandnames(self):
        self._bandnames = OrderedDict(('B%s' % band, i) for i, band in enumerate(range(1, self.bands + 1)))
        if self._metadata is not None:
213
            self.metadata.band_meta['band_names'] = list(self._bandnames.keys())
214

215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254
    @property
    def is_inmem(self):
        """Check if associated image array is completely loaded into memory."""
        return isinstance(self.arr, np.ndarray)

    @property
    def shape(self):
        """Get the array shape of the associated image array."""
        if self.is_inmem:
            return self.arr.shape
        else:
            if self._shape:
                return self._shape
            else:
                self.set_gdalDataset_meta()
                return self._shape

    @property
    def ndim(self):
        """Get the number dimensions of the associated image array."""
        return len(self.shape)

    @property
    def rows(self):
        """Get the number of rows of the associated image array."""

        return self.shape[0]

    @property
    def columns(self):
        """Get the number of columns of the associated image array."""

        return self.shape[1]

    cols = alias_property('columns')

    @property
    def bands(self):
        """Get the number of bands of the associated image array."""

255
        return self.shape[2] if len(self.shape) > 2 else 1
256 257 258 259 260 261 262 263 264 265 266 267 268 269 270

    @property
    def dtype(self):
        """Get the numpy data type of the associated image array."""

        if self._dtype:
            return self._dtype
        elif self.is_inmem:
            return self.arr.dtype
        else:
            self.set_gdalDataset_meta()
            return self._dtype

    @property
    def geotransform(self):
271
        """Get the GDAL GeoTransform of the associated image, e.g., (283500.0, 5.0, 0.0, 4464500.0, 0.0, -5.0)"""
272 273 274 275 276 277 278

        if self._geotransform:
            return self._geotransform
        elif not self.is_inmem:
            self.set_gdalDataset_meta()
            return self._geotransform
        else:
279
            return [0, 1, 0, 0, 0, -1]
280 281 282

    @geotransform.setter
    def geotransform(self, gt):
283
        # type: (Union[list, tuple]) -> None
284 285
        assert isinstance(gt, (list, tuple)) and len(gt) == 6,\
            'geotransform must be a list with 6 numbers. Got %s.' % str(gt)
286

287
        for i in gt:
288
            assert is_number(i), "geotransform must contain only numbers. Got '%s' (type: %s)." % (i, type(i))
289

290
        self._geotransform = gt
291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312

    gt = alias_property('geotransform')

    @property
    def xgsd(self):
        """Get the X resolution in units of the given or detected projection."""

        return self.geotransform[1]

    @property
    def ygsd(self):
        """Get the Y resolution in units of the given or detected projection."""

        return abs(self.geotransform[5])

    @property
    def xygrid_specs(self):
        """
        Get the specifications for the X/Y coordinate grid, e.g. [[15,30], [0,30]] for a coordinate with its origin
        at X/Y[15,0] and a GSD of X/Y[15,30].
        """

313
        def get_grid(gt, xgsd, ygsd): return [[gt[0], gt[0] + xgsd], [gt[3], gt[3] - ygsd]]
314 315 316 317 318 319 320 321 322 323 324 325 326
        return get_grid(self.geotransform, self.xgsd, self.ygsd)

    @property
    def projection(self):
        """
        Get the projection of the associated image. Setting the projection is only allowed if GeoArray has been
        instanced from memory or the associated file on disk has no projection.
        """

        if self._projection:
            return self._projection
        elif not self.is_inmem:
            self.set_gdalDataset_meta()
327
            return self._projection  # or "LOCAL_CS[\"MAP\"]"
328
        else:
329
            return ''  # '"LOCAL_CS[\"MAP\"]"
330 331 332

    @projection.setter
    def projection(self, prj):
333
        # type: (str) -> None
334
        if self.filePath and self.projection:
335
            assert self.projection is None or prj_equal(self.projection, prj), \
336
                "Cannot set %s.projection to the given value because it does not match the projection from the file " \
337
                "on disk." % self.__class__.__name__
338 339 340 341 342 343 344 345 346 347 348 349 350
        else:
            self._projection = prj

    prj = alias_property('projection')

    @property
    def epsg(self):
        """Get the EPSG code of the projection of the GeoArray."""

        return WKT2EPSG(self.projection)

    @epsg.setter
    def epsg(self, epsg_code):
351
        # type: (int) -> None
352 353 354 355 356 357 358
        self.projection = EPSG2WKT(epsg_code)

    @property
    def box(self):
        mapPoly = get_footprint_polygon(get_corner_coordinates(gt=self.geotransform, cols=self.columns, rows=self.rows))
        return boxObj(gt=self.geotransform, prj=self.projection, mapPoly=mapPoly)

359 360 361 362 363 364 365 366
    @property
    def is_map_geo(self):
        # type: () -> bool
        """
        Returns 'True' if the GeoArray instance has a valid geoinformation with map instead of image coordinates.
        """
        return self.gt and list(self.gt) != [0, 1, 0, 0, 0, -1] and self.prj

367 368 369 370 371 372 373 374 375 376 377 378 379 380 381
    @property
    def nodata(self):
        """
        Get the nodata value of the GeoArray. If GeoArray has been instanced with a file path the file is checked
        for an existing nodata value. Otherwise (if no value is exlicitly given during object instanciation) the nodata
        value is tried to be automatically detected.
        """

        if self._nodata is not None:
            return self._nodata
        else:
            # try to get nodata value from file
            if not self.is_inmem:
                self.set_gdalDataset_meta()
            if self._nodata is None:
382
                self.find_noDataVal()
383 384 385 386 387 388
                if self._nodata == 'ambiguous':
                    warnings.warn('Nodata value could not be clearly identified. It has been set to None.')
                    self._nodata = None
                else:
                    if self._nodata is not None and not self.q:
                        print("Automatically detected nodata value for %s '%s': %s"
389
                              % (self.__class__.__name__, self.basename, self._nodata))
390 391 392 393
            return self._nodata

    @nodata.setter
    def nodata(self, value):
394
        # type: (Union[int, None]) -> None
395 396
        self._nodata = value

397 398 399
        if self._metadata and value is not None:
            self.metadata.global_meta.update({'data ignore value': str(value)})

400 401 402 403 404 405 406 407 408
    @property
    def mask_nodata(self):
        """
        Get the nodata mask of the associated image array. It is calculated using all image bands.
        """

        if self._mask_nodata is not None:
            return self._mask_nodata
        else:
409
            self.calc_mask_nodata()  # sets self._mask_nodata
410 411 412 413 414 415 416 417 418 419 420
            return self._mask_nodata

    @mask_nodata.setter
    def mask_nodata(self, mask):
        """Set bad data mask.

        :param mask:    Can be a file path, a numpy array or an instance o GeoArray.
        """

        if mask is not None:
            from .masks import NoDataMask
421 422
            geoArr_mask = NoDataMask(mask, progress=self.progress, q=self.q)
            geoArr_mask.gt = geoArr_mask.gt if geoArr_mask.gt not in [None, [0, 1, 0, 0, 0, -1]] else self.gt
423
            geoArr_mask.prj = geoArr_mask.prj if geoArr_mask.prj else self.prj
424
            imName = "the %s '%s'" % (self.__class__.__name__, self.basename)
425 426 427 428

            assert geoArr_mask.bands == 1, \
                'Expected one single band as nodata mask for %s. Got %s bands.' % (self.basename, geoArr_mask.bands)
            assert geoArr_mask.shape[:2] == self.shape[:2], 'The provided nodata mask must have the same number of ' \
429
                                                            'rows and columns as the %s itself.' % imName
430 431
            assert geoArr_mask.gt == self.gt, \
                'The geotransform of the given nodata mask for %s must match the geotransform of the %s itself. ' \
432
                'Got %s.' % (imName, self.__class__.__name__, geoArr_mask.gt)
433 434
            assert not geoArr_mask.prj or prj_equal(geoArr_mask.prj, self.prj), \
                'The projection of the given nodata mask for the %s must match the projection of the %s itself.' \
435
                % (imName, self.__class__.__name__)
436 437

            self._mask_nodata = geoArr_mask
438 439 440 441 442 443
        else:
            del self.mask_nodata

    @mask_nodata.deleter
    def mask_nodata(self):
        self._mask_nodata = None
444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462

    @property
    def mask_baddata(self):
        """
        Returns the bad data mask for the associated image array if it has been explicitly previously. It can be set
         by passing a file path, a numpy array or an instance of GeoArray to the setter of this property.
        """

        return self._mask_baddata

    @mask_baddata.setter
    def mask_baddata(self, mask):
        """Set bad data mask.

        :param mask:    Can be a file path, a numpy array or an instance o GeoArray.
        """

        if mask is not None:
            from .masks import BadDataMask
463 464
            geoArr_mask = BadDataMask(mask, progress=self.progress, q=self.q)
            geoArr_mask.gt = geoArr_mask.gt if geoArr_mask.gt not in [None, [0, 1, 0, 0, 0, -1]] else self.gt
465
            geoArr_mask.prj = geoArr_mask.prj if geoArr_mask.prj else self.prj
466
            imName = "the %s '%s'" % (self.__class__.__name__, self.basename)
467 468 469 470

            assert geoArr_mask.bands == 1, \
                'Expected one single band as bad data mask for %s. Got %s bands.' % (self.basename, geoArr_mask.bands)
            assert geoArr_mask.shape[:2] == self.shape[:2], 'The provided bad data mask must have the same number of ' \
471
                                                            'rows and columns as the %s itself.' % imName
472 473
            assert geoArr_mask.gt == self.gt, \
                'The geotransform of the given bad data mask for %s must match the geotransform of the %s itself. ' \
474
                'Got %s.' % (imName, self.__class__.__name__, geoArr_mask.gt)
475 476
            assert prj_equal(geoArr_mask.prj, self.prj), \
                'The projection of the given bad data mask for the %s must match the projection of the %s itself.' \
477
                % (imName, self.__class__.__name__)
478 479

            self._mask_baddata = geoArr_mask
480 481 482 483 484 485
        else:
            del self.mask_baddata

    @mask_baddata.deleter
    def mask_baddata(self):
        self._mask_baddata = None
486 487 488 489 490 491 492 493 494 495

    @property
    def footprint_poly(self):
        # FIXME should return polygon in image coordinates if no projection is available
        """
        Get the footprint polygon of the associated image array (returns an instance of shapely.geometry.Polygon.
        """

        if self._footprint_poly is None:
            assert self.mask_nodata is not None, 'A nodata mask is needed for calculating the footprint polygon. '
496
            if False not in self.mask_nodata[:]:
497 498 499 500
                # do not run raster2polygon if whole image is filled with data
                self._footprint_poly = self.box.mapPoly
            else:
                try:
501 502
                    multipolygon = raster2polygon(self.mask_nodata.astype(np.uint8), self.gt, self.prj, exact=False,
                                                  progress=self.progress, q=self.q, maxfeatCount=10, timeout=3)
503
                    self._footprint_poly = fill_holes_within_poly(multipolygon)
504
                except (RuntimeError, TimeoutError):
505 506 507 508
                    if not self.q:
                        warnings.warn("\nCalculation of footprint polygon failed for %s '%s'. Using outer bounds. One "
                                      "reason could be that the nodata value appears within the actual image (not only "
                                      "as fill value). To avoid this use another nodata value. Current nodata value is "
509
                                      "%s." % (self.__class__.__name__, self.basename, self.nodata))
510 511 512
                    self._footprint_poly = self.box.mapPoly

            # validation
513 514 515 516
            assert not polyVertices_outside_poly(self._footprint_poly, self.box.mapPoly), \
                "Computing footprint polygon for %s '%s' failed. The resulting polygon is partly or completely " \
                "outside of the image bounds." % (self.__class__.__name__, self.basename)
            # assert self._footprint_poly
517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534
            # for XY in self.corner_coord:
            #    assert self.GeoArray.box.mapPoly.contains(Point(XY)) or self.GeoArray.box.mapPoly.touches(Point(XY)), \
            #        "The corner position '%s' is outside of the %s." % (XY, self.imName)

        return self._footprint_poly

    @footprint_poly.setter
    def footprint_poly(self, poly):
        if isinstance(poly, Polygon):
            self._footprint_poly = poly
        elif isinstance(poly, str):
            self._footprint_poly = shply_loads(poly)
        else:
            raise ValueError("'footprint_poly' can only be set from a shapely polygon or a WKT string.")

    @property
    def metadata(self):
        """
535
        Returns a DataFrame containing all available metadata (read from file if available).
536 537 538 539
        Use 'metadata[band_index].to_dict()' to get a metadata dictionary for a specific band.
        Use 'metadata.loc[row_name].to_dict()' to get all metadata values of the same key for all bands as dictionary.
        Use 'metadata.loc[row_name, band_index] = value' to set a new value.

540
        :return:  pandas.DataFrame
541 542 543 544 545
        """

        if self._metadata is not None:
            return self._metadata
        else:
546
            default = GDAL_Metadata(nbands=self.bands, nodata_allbands=self._nodata)
547

548 549 550 551 552 553 554 555
            self._metadata = default
            if not self.is_inmem:
                self.set_gdalDataset_meta()
                return self._metadata
            else:
                return self._metadata

    @metadata.setter
556 557 558 559 560 561
    def metadata(self, meta):
        if not isinstance(meta, GDAL_Metadata) or meta.bands != self.bands:
            raise ValueError("%s.metadata can only be set with an instance of geoarray.metadata.GDAL_Metadata of "
                             "which the band number corresponds to the band number of %s."
                             % (self.__class__.__name__, self.__class__.__name__))
        self._metadata = meta
562 563 564 565

    meta = alias_property('metadata')

    def __getitem__(self, given):
566
        if isinstance(given, (int, float, slice, np.integer, np.floating)) and self.ndim == 3:
567 568 569 570 571 572 573 574 575 576 577 578 579
            # handle 'given' as index for 3rd (bands) dimension
            if self.is_inmem:
                return self.arr[:, :, given]
            else:
                return self.from_path(self.arg, [given])

        elif isinstance(given, str):
            # behave like a dictionary and return the corresponding band
            if self.bandnames:
                if given not in self.bandnames:
                    raise ValueError("'%s' is not a known band. Known bands are: %s"
                                     % (given, ', '.join(list(self.bandnames.keys()))))
                if self.is_inmem:
580
                    return self.arr if self.ndim == 2 else self.arr[:, :, self.bandnames[given]]
581 582 583 584
                else:
                    return self.from_path(self.arg, [self.bandnames[given]])
            else:
                raise ValueError('String indices are only supported if %s has been instanced with bandnames given.'
585
                                 % self.__class__.__name__)
586 587 588 589 590 591 592

        elif isinstance(given, (tuple, list)):
            # handle requests like geoArr[[1,2],[3,4]  -> not implemented in from_path if array is not in mem
            types = [type(i) for i in given]
            if list in types or tuple in types:
                self.to_mem()

593
            if len(given) == 3:
594 595

                # handle strings in the 3rd dim of 'given' -> convert them to a band index
596
                if isinstance(given[2], str):
597 598 599 600 601 602 603 604
                    if self.bandnames:
                        if given[2] not in self.bandnames:
                            raise ValueError("'%s' is not a known band. Known bands are: %s"
                                             % (given[2], ', '.join(list(self.bandnames.keys()))))

                        band_idx = self.bandnames[given[2]]
                        # NOTE: the string in the 3rd is ignored if ndim==2 and band_idx==0
                        if self.is_inmem:
605
                            return self.arr if (self.ndim == 2 and band_idx == 0) else self.arr[:, :, band_idx]
606
                        else:
607 608
                            getitem_params = \
                                given[:2] if (self.ndim == 2 and band_idx == 0) else given[:2] + (band_idx,)
609 610 611 612 613 614 615
                            return self.from_path(self.arg, getitem_params)
                    else:
                        raise ValueError(
                            'String indices are only supported if %s has been instanced with bandnames given.'
                            % self.__class__.__name__)

                # in case a third dim is requested from 2D-array -> ignore 3rd dim if 3rd dim is 0
616
                elif self.ndim == 2 and given[2] == 0:
617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640
                    if self.is_inmem:
                        return self.arr[given[:2]]
                    else:
                        return self.from_path(self.arg, given[:2])

        # if nothing has been returned until here -> behave like a numpy array
        if self.is_inmem:
            return self.arr[given]
        else:
            getitem_params = [given] if isinstance(given, slice) else given
            return self.from_path(self.arg, getitem_params)

    def __setitem__(self, idx, array2set):
        """Overwrites the pixel values of GeoArray.arr with the given array.

        :param idx:         <int, list, slice> the index position to overwrite
        :param array2set:   <np.ndarray> array to be set. Must be compatible to the given index position.
        :return:
        """

        if self.is_inmem:
            self.arr[idx] = array2set
        else:
            raise NotImplementedError('Item assignment for %s instances that are not in memory is not yet supported.'
641
                                      % self.__class__.__name__)
642 643 644

    def __getattr__(self, attr):
        # check if the requested attribute can not be present because GeoArray has been instanced with an array
645 646
        attrsNot2Link2np = ['__deepcopy__']   # attributes we don't want to inherit from numpy.ndarray

647 648
        if attr not in self.__dir__() and not self.is_inmem and attr in ['shape', 'dtype', 'geotransform',
                                                                         'projection']:
649 650
            self.set_gdalDataset_meta()

651 652
        if attr in self.__dir__():  # __dir__() includes also methods and properties
            return self.__getattribute__(attr)  # __getattribute__ avoids infinite loop
653
        elif attr not in attrsNot2Link2np and hasattr(np.array([]), attr):
654 655
            return self[:].__getattribute__(attr)
        else:
656
            raise AttributeError("%s object has no attribute '%s'." % (self.__class__.__name__, attr))
657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676

    def __getstate__(self):
        """Defines how the attributes of GMS object are pickled."""

        # clean array cache in order to avoid cache pickling
        self.flush_cache()

        return self.__dict__

    def __setstate__(self, state):
        """Defines how the attributes of GMS object are unpickled.
        NOTE: This method has been implemented because otherwise pickled and unpickled instances show recursion errors
        within __getattr__ when requesting any attribute.
        """

        self.__dict__ = state

    def calc_mask_nodata(self, fromBand=None, overwrite=False):
        """Calculates a no data mask with (values: 0=nodata; 1=data)

677 678 679
        NOTE:   Only pixel containing the nodata values in ALL bands are recognized as nodata pixel. If they contain a
                pixel value different from the nodata value in any band they are good data pixels.

680 681 682 683 684 685
        :param fromBand:  <int> index of the band to be used (if None, all bands are used)
        :param overwrite: <bool> whether to overwrite existing nodata mask that has already been calculated
        :return:
        """
        if self._mask_nodata is None or overwrite:
            assert self.ndim in [2, 3], "Only 2D or 3D arrays are supported. Got a %sD array." % self.ndim
686
            arr = self[:, :, fromBand] if self.ndim == 3 and fromBand is not None else self[:]
687

688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703
            if self.nodata is None:
                mask = np.ones((self.rows, self.cols), np.bool)

            elif np.isnan(self.nodata):
                nanmask = np.isnan(arr)
                nanbands = np.all(np.all(nanmask, axis=0), axis=0)

                if np.all(nanbands):
                    mask = np.full(arr.shape[:2], False)
                elif arr.ndim == 2:
                    mask = ~np.isnan(arr)
                else:
                    idx_1st_databand = np.argwhere(~nanbands)[0][0]
                    mask = ~np.isnan(arr[:, :, idx_1st_databand])
                    mask[~mask] = np.any(~np.isnan(arr[~mask]), axis=1)

704
            else:
705 706 707 708 709 710
                bandmeans = np.mean(np.mean(arr, axis=0), axis=0)

                if np.nanmean(bandmeans) == self.nodata:
                    mask = np.full(arr.shape[:2], False)
                elif arr.ndim == 2:
                    mask = arr != self.nodata
711
                else:
712 713 714 715 716 717 718
                    idx_1st_databand = np.argwhere(bandmeans != self.nodata)[0][0]
                    mask = np.array(arr[:, :, idx_1st_databand] != self.nodata)
                    mask[~mask] = np.any(arr[~mask] != self.nodata, axis=1)

            self.mask_nodata = mask

            return mask
719

720 721 722 723 724 725 726 727
    def find_noDataVal(self, bandIdx=0, sz=3):
        """Tries to derive no data value from homogenious corner pixels within 3x3 windows (by default).
        :param bandIdx:
        :param sz: window size in which corner pixels are analysed
        """
        wins = [self[0:sz, 0:sz, bandIdx], self[0:sz, -sz:, bandIdx],
                self[-sz:, -sz:, bandIdx], self[-sz:, 0:sz, bandIdx]]  # UL, UR, LR, LL

728 729
        means, stds = [np.mean(win) for win in wins], [np.std(win) for win in wins]
        possVals = [mean for mean, std in zip(means, stds) if std == 0 or np.isnan(std)]
730 731 732 733
        # possVals==[]: all corners are filled with data; np.std(possVals)==0: noDataVal clearly identified

        if possVals:
            if np.std(possVals) != 0:
734 735 736 737 738 739
                if np.isnan(np.std(possVals)):
                    # at least one of the possible values is np.nan
                    nodata = np.nan
                else:
                    # different possible nodata values have been found in the image corner
                    nodata = 'ambiguous'
740 741 742 743 744
            else:
                if len(possVals) <= 2:
                    # each window in each corner
                    warnings.warn("\nAutomatic nodata value detection returned the value %s for GeoArray '%s' but this "
                                  "seems to be unreliable (occurs in only %s). To avoid automatic detection, just pass "
745 746 747
                                  "the correct nodata value."
                                  % (possVals[0], self.basename, ('2 image corners' if len(possVals) == 2 else
                                                                  '1 image corner')))
748
                nodata = possVals[0]
749
        else:
750 751
            nodata = None

752
        self.nodata = nodata
753
        return nodata
754

755 756 757 758 759 760 761 762 763 764
    def set_gdalDataset_meta(self):
        """Retrieves GDAL metadata from file. This function is only executed once to avoid overwriting of user defined
         attributes, that are defined after object instanciation.

        :return:
        """

        if not self._gdalDataset_meta_already_set:
            assert self.filePath
            ds = gdal.Open(self.filePath)
765 766 767
            if not ds:
                raise Exception('Error reading file:  ' + gdal.GetLastErrorMsg())

768
            # set private class variables (in order to avoid recursion error)
769 770
            self._shape = tuple([ds.RasterYSize, ds.RasterXSize] + ([ds.RasterCount] if ds.RasterCount > 1 else []))
            self._dtype = gdal_array.GDALTypeCodeToNumericTypeCode(ds.GetRasterBand(1).DataType)
771
            self._geotransform = list(ds.GetGeoTransform())
772 773

            # for some reason GDAL reads arbitrary geotransforms as (0, 1, 0, 0, 0, 1) instead of (0, 1, 0, 0, 0, -1)
774
            self._geotransform[5] = -abs(self._geotransform[5])  # => force ygsd to be negative
775

776 777
            # temp conversion to EPSG needed because GDAL seems to modify WKT string when writing file to disk
            # (e.g. using gdal_merge) -> conversion to EPSG and back undos that
778 779
            wkt = ds.GetProjection()
            self._projection = EPSG2WKT(WKT2EPSG(wkt)) if not isLocal(wkt) else ''
780

781 782 783
            if 'nodata' not in self._initParams or self._initParams['nodata'] is None:
                band = ds.GetRasterBand(1)
                # FIXME this does not support different nodata values within the same file
784
                self.nodata = band.GetNoDataValue()
785

786 787 788
            # set metadata attribute
            if self.is_inmem or not self.filePath:
                # metadata cannot be read from disk -> set it to the default
789
                self._metadata = GDAL_Metadata(nbands=self.bands, nodata_allbands=self._nodata)
790

791 792
            else:
                self._metadata = GDAL_Metadata(filePath=self.filePath)
793

794 795 796 797 798 799
            # copy over the band names
            if 'band_names' in self.metadata.band_meta and self.metadata.band_meta['band_names']:
                self.bandnames = self.metadata.band_meta['band_names']

            # noinspection PyUnusedLocal
            ds = None
800 801 802 803 804 805 806 807 808 809 810 811 812 813

        self._gdalDataset_meta_already_set = True

    def from_path(self, path, getitem_params=None):
        # type: (str, list) -> np.ndarray
        """Read a GDAL compatible raster image from disk, with respect to the given image position.
        NOTE: If the requested array position is already in cache, it is returned from there.

        :param path:            <str> the file path of the image to read
        :param getitem_params:  <list> a list of slices in the form [row_slice, col_slice, band_slice]
        :return out_arr:        <np.ndarray> the output array
        """

        ds = gdal.Open(path)
814 815 816
        if not ds:
            raise Exception('Error reading file:  ' + gdal.GetLastErrorMsg())

817
        R, C, B = ds.RasterYSize, ds.RasterXSize, ds.RasterCount
818
        del ds
819

820
        # convert getitem_params to subset area to be read #
821 822 823 824
        rS, rE, cS, cE, bS, bE, bL = [None] * 7

        # populate rS, rE, cS, cE, bS, bE, bL
        if getitem_params:
825
            # populate rS, rE, cS, cE
826 827 828 829 830
            if len(getitem_params) >= 2:
                givenR, givenC = getitem_params[:2]
                if isinstance(givenR, slice):
                    rS = givenR.start
                    rE = givenR.stop - 1 if givenR.stop is not None else None
831
                elif isinstance(givenR, (int, np.integer)):
832 833 834 835 836
                    rS = givenR
                    rE = givenR
                if isinstance(givenC, slice):
                    cS = givenC.start
                    cE = givenC.stop - 1 if givenC.stop is not None else None
837
                elif isinstance(givenC, (int, np.integer)):
838 839
                    cS = givenC
                    cE = givenC
840 841

            # populate bS, bE, bL
842 843 844 845 846
            if len(getitem_params) in [1, 3]:
                givenB = getitem_params[2] if len(getitem_params) == 3 else getitem_params[0]
                if isinstance(givenB, slice):
                    bS = givenB.start
                    bE = givenB.stop - 1 if givenB.stop is not None else None
847
                elif isinstance(givenB, (int, np.integer)):
848 849
                    bS = givenB
                    bE = givenB
850
                elif isinstance(givenB, (tuple, list)):
851 852 853
                    typesInGivenB = [type(i) for i in givenB]
                    assert len(list(set(typesInGivenB))) == 1, \
                        'Mixed data types within the list of bands are not supported.'
854
                    if isinstance(givenB[0], (int, np.integer)):
855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876
                        bL = list(givenB)
                    elif isinstance(givenB[0], str):
                        bL = [self.bandnames[i] for i in givenB]
                elif type(givenB) in [str]:
                    bL = [self.bandnames[givenB]]

        # set defaults for not given values
        rS = rS if rS is not None else 0
        rE = rE if rE is not None else R - 1
        cS = cS if cS is not None else 0
        cE = cE if cE is not None else C - 1
        bS = bS if bS is not None else 0
        bE = bE if bE is not None else B - 1
        bL = list(range(bS, bE + 1)) if not bL else bL

        # convert negative to positive ones
        rS = rS if rS >= 0 else self.rows + rS
        rE = rE if rE >= 0 else self.rows + rE
        cS = cS if cS >= 0 else self.columns + cS
        cE = cE if cE >= 0 else self.columns + cE
        bS = bS if bS >= 0 else self.bands + bS
        bE = bE if bE >= 0 else self.bands + bE
877
        bL = [b if b >= 0 else (self.bands + b) for b in bL]
878 879

        # validate subset area bounds to be read
880 881 882 883 884 885 886
        def msg(v, idx, sz):
            # FIXME numpy raises that error ONLY for the 2nd axis
            return '%s is out of bounds for axis %s with size %s' % (v, idx, sz)

        for val, axIdx, axSize in zip([rS, rE, cS, cE, bS, bE], [0, 0, 1, 1, 2, 2], [R, R, C, C, B, B]):
            if not 0 <= val <= axSize - 1:
                raise ValueError(msg(val, axIdx, axSize))
887 888

        # summarize requested array position in arr_pos
889
        # NOTE: # bandlist must be string because truth value of an array with more than one element is ambiguous
890 891
        arr_pos = dict(rS=rS, rE=rE, cS=cS, cE=cE, bS=bS, bE=bE, bL=bL)

892 893
        def _ensure_np_shape_consistency_3D_2D(arr):
            """Ensure numpy output shape consistency according to the given indexing parameters.
894 895 896 897 898 899 900 901

            This may require 3D to 2D conversion in case out_arr can be represented by a 2D array AND index has been
            provided as integer (avoids shapes like (1,2,2). It also may require 2D to 3D conversion in case only one
            band has been requested and the 3rd dimension has been provided as a slice.

            NOTE: -> numpy also returns a 2D array in that case
            NOTE: if array is indexed with a slice -> keep it a 3D array
            """
902 903 904 905
            # a single value -> return as float/int
            if arr.ndim == 2 and arr.size == 1:
                arr = arr[0, 0]

906 907 908 909 910 911
            # 2D -> 3D
            if arr.ndim == 2 and isinstance(getitem_params, (tuple, list)) and len(getitem_params) == 3 and \
                    isinstance(getitem_params[2], slice):
                arr = arr[:, :, np.newaxis]

            # 3D -> 2D
Daniel Scheffler's avatar
Bugfix.  
Daniel Scheffler committed
912
            if 1 in arr.shape and len(getitem_params) != 1:
913 914
                outshape = []
                for i, sh in enumerate(arr.shape):
915
                    if sh == 1 and isinstance(getitem_params[i], (int, np.integer, float, np.floating)):
916 917 918 919 920 921 922 923
                        pass
                    else:
                        outshape.append(sh)

                arr = arr.reshape(*outshape)

            return arr

924
        # check if the requested array position is already in cache -> if yes, return it from there
925
        if self._arr_cache is not None and self._arr_cache['pos'] == arr_pos:
926
            out_arr = self._arr_cache['arr_cached']
927
            out_arr = _ensure_np_shape_consistency_3D_2D(out_arr)
928 929 930 931 932 933 934

        else:
            # TODO insert a multiprocessing.Lock here in order to prevent IO bottlenecks?
            # read subset area from disk
            if bL == list(range(0, B)):
                tempArr = gdalnumeric.LoadFile(path, cS, rS, cE - cS + 1, rE - rS + 1)
                out_arr = np.swapaxes(np.swapaxes(tempArr, 0, 2), 0, 1) if B > 1 else tempArr
935 936
                if out_arr is None:
                    raise Exception('Error reading file:  ' + gdal.GetLastErrorMsg())
937 938 939 940
            else:
                ds = gdal.Open(path)
                if len(bL) == 1:
                    band = ds.GetRasterBand(bL[0] + 1)
941
                    out_arr = band.ReadAsArray(cS, rS, cE - cS + 1, rE - rS + 1)
942 943
                    if out_arr is None:
                        raise Exception('Error reading file:  ' + gdal.GetLastErrorMsg())
944
                    del band
945 946 947 948 949
                else:
                    out_arr = np.empty((rE - rS + 1, cE - cS + 1, len(bL)))
                    for i, bIdx in enumerate(bL):
                        band = ds.GetRasterBand(bIdx + 1)
                        out_arr[:, :, i] = band.ReadAsArray(cS, rS, cE - cS + 1, rE - rS + 1)
950 951
                        if out_arr is None:
                            raise Exception('Error reading file:  ' + gdal.GetLastErrorMsg())
952
                        del band
953

954
                del ds
955

956
            out_arr = _ensure_np_shape_consistency_3D_2D(out_arr)
957

958
            # only set self.arr if the whole cube has been read (in order to avoid sudden shape changes)
959
            if out_arr.shape == self.shape:
960 961 962 963 964
                self.arr = out_arr

            # write _arr_cache
            self._arr_cache = dict(pos=arr_pos, arr_cached=out_arr)

965 966
        return out_arr  # TODO implement check of returned datatype (e.g. NoDataMask should always return np.bool
        # TODO -> would be np.int8 if an int8 file is read from disk
967 968 969 970 971 972 973 974

    def save(self, out_path, fmt='ENVI', creationOptions=None):
        # type: (str, str, list) -> None
        """Write the raster data to disk.

        :param out_path:        <str> output path
        :param fmt:             <str> the output format / GDAL driver code to be used for output creation, e.g. 'ENVI'
                                Refer to http://www.gdal.org/formats_list.html to get a full list of supported formats.
975 976
        :param creationOptions: <list> GDAL creation options,
                                e.g., ["QUALITY=80", "REVERSIBLE=YES", "WRITE_METADATA=YES"]
977 978 979
        """

        if not self.q:
980 981
            print('Writing GeoArray of size %s to %s.' % (self.shape, out_path))
        assert self.ndim in [2, 3], 'Only 2D- or 3D arrays are supported.'
982 983 984 985 986 987 988 989 990

        driver = gdal.GetDriverByName(fmt)
        if driver is None:
            raise Exception("'%s' is not a supported GDAL driver. Refer to www.gdal.org/formats_list.html for full "
                            "list of GDAL driver codes." % fmt)

        if not os.path.isdir(os.path.dirname(out_path)):
            os.makedirs(os.path.dirname(out_path))

991 992
        envi_metadict = self.metadata.to_ENVI_metadict()

993
        if self.is_inmem:
Daniel Scheffler's avatar
Daniel Scheffler committed
994 995 996 997 998
            ds_inmem = get_GDAL_ds_inmem(self.arr, self.geotransform, self.projection,
                                         self.nodata)  # expects rows,columns,bands

            # write dataset
            ds_out = driver.CreateCopy(out_path, ds_inmem, options=creationOptions if creationOptions else [])
999 1000 1001 1002 1003 1004

            # # rows, columns, bands => bands, rows, columns
            # out_arr = self.arr if self.ndim == 2 else np.swapaxes(np.swapaxes(self.arr, 0, 2), 1, 2)
            # gdalnumeric.SaveArray(out_arr, out_path, format=fmt, prototype=ds_inmem)  # expects bands,rows,columns
            # ds_out = gdal.Open(out_path)

Daniel Scheffler's avatar
Daniel Scheffler committed
1005
            del ds_inmem
1006

1007 1008 1009 1010
            ################
            # set metadata #
            ################

Daniel Scheffler's avatar
Daniel Scheffler committed
1011
            # NOTE:  The dataset has to be written BEFORE metadata are added. Otherwise, metadata are not written.
1012 1013 1014

            # ENVI #
            ########
1015 1016
            if fmt == 'ENVI':
                ds_out.SetMetadata(envi_metadict, 'ENVI')
1017

1018
                if 'band_names' in envi_metadict:
1019 1020
                    for bidx in range(self.bands):
                        band = ds_out.GetRasterBand(bidx + 1)
1021 1022 1023 1024
                        bandname = self.metadata.band_meta['band_names'][bidx].strip()
                        band.SetDescription(bandname)

                        assert band.GetDescription() == bandname
1025 1026 1027 1028
                        del band

                if 'description' in envi_metadict:
                    ds_out.SetDescription(envi_metadict['description'])
1029

1030 1031
                ds_out.FlushCache()
                gdal.Unlink(out_path + '.aux.xml')
1032

1033
            elif self.metadata.all_meta: