def crowns_to_polys_raster(self):
''' Converts tree crown raster to individual polygons and stores them
in the tree dataframe
'''
polys = []
for feature in rioshapes(self.crowns, mask=self.crowns.astype(bool)):
# Convert pixel coordinates to lon/lat
edges = feature[0]['coordinates'][0].copy()
for i in range(len(edges)):
edges[i] = self._to_lonlat(*edges[i], self.resolution)
# poly_smooth = self.smooth_poly(Polygon(edges), s=None, k=9)
polys.append(Polygon(edges))
self.trees.crown_poly_raster = polys
def crowns_to_polys_smooth(self, store_las=True):
""" Smooth crown polygons using Dalponte & Coomes (2016) approach:
Builds a convex hull around first return points (which lie within the
rasterized crowns).
Optionally, the trees in the LiDAR point cloud are classified based on
the generated convex hull.
Parameters
----------
store_las : bool
set to True if LiDAR point clouds shopuld be classified
and stored externally
"""
print('Converting LAS point cloud to shapely points')
geometry = [Point(xy) for xy in zip(self.las.x, self.las.y)]
lidar_geodf = gpd.GeoDataFrame(self.las,
crs=f'epsg:{self.epsg}',
geometry=geometry)
print('Converting raster crowns to shapely polygons')
polys = []
for feature in rioshapes(self.crowns, mask=self.crowns.astype(bool)):
edges = np.array(list(zip(*feature[0]['coordinates'][0])))
edges = np.array(
self._to_lonlat(edges[0], edges[1], self.resolution)).T
polys.append(Polygon(edges))
crown_geodf = gpd.GeoDataFrame(pd.DataFrame(np.arange(len(
self.trees))),
crs=f'epsg:{self.epsg}',
geometry=polys)
print('Attach LiDAR points to corresponding crowns')
lidar_in_crowns = gpd.sjoin(lidar_geodf,
crown_geodf,
op='within',
how="inner")
lidar_tree_class = np.zeros(lidar_in_crowns['index_right'].size)
lidar_tree_mask = np.zeros(lidar_in_crowns['index_right'].size,
dtype=bool)
print('Create convex hull around first return points')
polys = []
for tidx in range(len(self.trees)):
bool_indices = lidar_in_crowns['index_right'] == tidx
lidar_tree_class[bool_indices] = tidx
points = lidar_in_crowns[bool_indices]
# check that not all values are the same
if len(points.z) > 1 and not np.allclose(points.z,
points.iloc[0].z):
points = points[points.z >= threshold_otsu(points.z)]
points = points[points.return_num == 1]
hull = points.unary_union.convex_hull
polys.append(hull)
lidar_tree_mask[bool_indices] = \
lidar_in_crowns[bool_indices].within(hull)
self.trees.crown_poly_smooth = polys
if store_las:
print('Classifying point cloud')
lidar_in_crowns = lidar_in_crowns[lidar_tree_mask]
lidar_tree_class = lidar_tree_class[lidar_tree_mask]
header = laspy.header.Header()
self.outpath.mkdir(parents=True, exist_ok=True)
outfile = laspy.file.File(self.outpath / "trees.las",
mode="w",
header=header)
xmin = np.floor(np.min(lidar_in_crowns.x))
ymin = np.floor(np.min(lidar_in_crowns.y))
zmin = np.floor(np.min(lidar_in_crowns.z))
outfile.header.offset = [xmin, ymin, zmin]
outfile.header.scale = [0.001, 0.001, 0.001]
outfile.x = lidar_in_crowns.x
outfile.y = lidar_in_crowns.y
outfile.z = lidar_in_crowns.z
outfile.intensity = lidar_tree_class
outfile.close()
self.lidar_in_crowns = lidar_in_crowns
