def loadRaster(raster_file, md_dest=None):
'''
Load a raster dataset, and optionally reproject
Args:
raster_file: Path to a GeoTiff or .vrt file.
md_dest: A metadata file from sen2mosaic.Metadata().
Returns:
A numpy array
'''
# Load landcover map
ds_source = gdal.Open(raster_file, 0)
# If no reprojection required, return array
if md_dest is None:
return ds_source.GetRasterBand(1)
# Else reproject
else:
geo_t = ds_source.GetGeoTransform()
# Get extent and resolution of input raster
nrows = ds_source.RasterXSize
ncols = ds_source.RasterYSize
ulx = float(geo_t[0])
uly = float(geo_t[3])
xres = float(geo_t[1])
yres = float(geo_t[5])
lrx = ulx + (xres * ncols)
lry = uly + (yres * nrows)
extent = [ulx, lry, lrx, uly]
# Get EPSG
proj = ds_source.GetProjection()
srs = osr.SpatialReference(wkt=proj)
srs.AutoIdentifyEPSG()
EPSG = int(srs.GetAttrValue("AUTHORITY", 1))
# Add source metadata to a dictionary
md_source = sen2mosaic.Metadata(extent, xres, EPSG)
# Build an empty destination dataset
ds_dest = createGdalDataset(
md_dest,
nodata=ds_source.GetRasterBand(1).GetNoDataValue(),
dtype=1)
# And reproject landcover dataset to match input image
im_rep = np.squeeze(
_reprojectImage(ds_source, ds_dest, md_source, md_dest))
return im_rep
def main(source_files, target_extent, resolution, EPSG_code, training_data, forest_values, nonforest_values, level = '2A', field_name = '', n_processes = 1, max_images = 0, max_pixels = 5000, output_dir = os.getcwd(), output_name = 'S2', verbose = True):
'''main(source_files, training_data, target_extent, resolution, EPSG_code, n_processes = 1, max_pixels = 5000, output_dir = os.getcwd(), output_name = 'S2')
Extract pixel values from source_files and output as a np.savez() file. This is the function that is initiated from the command line.
Args:
source_files: A list of directories for Sentinel-2 input tiles.
target_extent: Extent of search area, in format [xmin, ymin, xmax, ymax]
resolution: Resolution to re-sample search area, in meters. Best to be 10 m, 20 m or 60 m to match Sentinel-2 resolution.
EPSG_code: EPSG code of search area.
training_data: A GeoTiff, .vrt of .shp file containing training pixels/polygons.
forest_values: A list of raster classes (integers) or shapefile attribute values (str) indicating forest in training_data.
nonforest_values: A list of raster classes (integers) or shapefile attribute values (str) indicating nonforest in training_data.
field_name: Shapefile attribute under which forest_values and nonforest_values can be found
n_processes: Number of processes, defaults to 1.
max_images: Maximum number of input tiles to extract data from. Defaults to 0, meaning all valid tiles.
max_pixels: Maximum number of pixels to extract for each class from each image. Defaults to 5000.
output_dir: Directory to output classifier predictors. Defaults to current working directory.
output_name: Name to precede output file. Defaults to 'S2'.
'''
assert type(n_processes) == int and n_processes > 0, "n_processes must be an integer > 0."
# Get output metadata
md_dest = sen2mosaic.Metadata(target_extent, resolution, EPSG_code)
# Load and sort input scenes
scenes = sen2mosaic.IO.loadSceneList(source_files, md_dest = md_dest, level = level, sort_by = 'date')#, verbose = verbose)
# Reduce number of inputs to max_images
if max_images > 0 and len(scenes) > max_images:
scenes = [scenes[i] for i in sorted(random.sample(range(len(scenes)), max_images))]
assert len(scenes) > 0, "No valid input files found at specified location."
# Extract pixel values
forest_px, nonforest_px = deforest.extraction.extractData(scenes, training_data, md_dest, forest_values, nonforest_values, field = field_name, subset = max_pixels, n_processes = n_processes, output = True, output_dir = output_dir, output_name = output_name, verbose = verbose)
def _extractData(input_list):
'''
Multiprocessing requires some gymnastics. This is a wrapper function to initiate extractData() for multiprocessing.
Args:
input_list: A list of inputs for a single source_file, in the format: [source_file. trainging_data, target_extent, resolution, EPSG_code, subset].
Returns:
A tuple with (a list of forest pixel values, a list of nonforest pixel values)
'''
source_file = input_list[0]
s2_res = input_list[1]
training_data = input_list[2]
target_extent = input_list[3]
resolution = input_list[4]
EPSG_code = input_list[5]
forest_values = input_list[6]
nonforest_values = input_list[7]
field_name = input_list[8]
subset = input_list[9]
verbose = input_list[10]
# Load input scene
md_dest = sen2mosaic.Metadata(target_extent, resolution, EPSG_code)
# Load scene with sen2mosaic
scene = sen2mosaic.core.LoadScene(source_file, resolution=s2_res)
return extractData([scene],
training_data,
md_dest,
forest_values,
nonforest_values,
field=field_name,
subset=subset)