def map_bands_to_indices(target_bands, satellite=8):
band_map = defaultdict(list)
for band in landsat_rasters()[satellite]:
band_map[band] = []
for band in static_rasters():
band_map[band] = []
for band in climate_rasters():
band_map[band] = []
image_directory = '/home/thomas/share/image_data/train/37_28_2013/'
extensions = (".tif", ".TIF")
for dirpath, dirnames, filenames in os.walk(image_directory):
for f in filenames:
if any(ext in f for ext in extensions):
for band in band_map:
if f.endswith(band):
band_map[band].append(os.path.join(dirpath, f))
for band in band_map:
band_map[band] = sorted(band_map[band]) # ensures ordering within bands - sort by time.
indices = []
i = 0
for feat in sorted(band_map.keys()): # ensures the stack is in the same order each time.
feature_rasters = band_map[feat]
for feature_raster in feature_rasters:
for band in target_bands:
if feature_raster.endswith(band):
indices.append(i)
i += 1
return indices
def median_of_three(paths_map, image_stack, target_shape, satellite=8):
j = 0
out_image_stack = np.zeros((19, target_shape[1], target_shape[2]))
out_idx = 0
for band in sorted(paths_map.keys()):
if band in landsat_rasters()[satellite]:
slc = np.zeros((3, target_shape[1], target_shape[2]))
for i, sub_band in enumerate(paths_map[band]):
slc[i] = image_stack[j]
j += 1
out_image_stack[out_idx] = np.median(slc, axis=0)
out_idx += 1
else:
out_image_stack[out_idx] = image_stack[j]
out_idx += 1
return out_image_stack
def _landsat_band_map(subdirectory, satellite=8):
band_map = dict()
for band in landsat_rasters()[satellite]:
band_map[band] = None
for band in static_rasters():
band_map[band] = None
for band in climate_rasters():
band_map[band] = None
extensions = (".tif", ".TIF")
for dirpath, dirnames, filenames in os.walk(subdirectory):
for f in filenames:
if any(ext in f for ext in extensions):
for band in band_map:
if f.endswith(band):
band_map[band] = os.path.join(dirpath, f)
return band_map
def mean_of_three(paths_map, image_stack, target_shape, satellite=8):
# iterate over paths_map
# iterate over each raster in paths_map
j = 0
out_image_stack = np.zeros((19, target_shape[1], target_shape[2]))
out_idx = 0
for band in sorted(paths_map.keys()):
if band in landsat_rasters()[satellite]:
for sub_band in paths_map[band]:
out_image_stack[out_idx] += image_stack[j]
j += 1
out_image_stack[out_idx] /= 3
out_idx += 1
else:
out_image_stack[out_idx] = image_stack[j]
out_idx += 1
return out_image_stack
def all_rasters(image_directory, satellite=8):
''' Recursively get all rasters in image_directory
and its subdirectories, and adds them to band_map. '''
band_map = defaultdict(list)
for band in landsat_rasters()[satellite]:
band_map[band] = []
for band in static_rasters():
band_map[band] = []
for band in climate_rasters():
band_map[band] = []
extensions = (".tif", ".TIF")
for dirpath, dirnames, filenames in os.walk(image_directory):
for f in filenames:
if any(ext in f for ext in extensions):
for band in band_map:
if f.endswith(band):
band_map[band].append(os.path.join(dirpath, f))
for band in band_map:
band_map[band] = sorted(band_map[band]) # ensures ordering within bands - sort by time.
return band_map