Spatial Agglomerative Cluster

The spatial_agglomerative_clustering() function segments a :class:DataCube by performing hierarchical agglomerative clustering on its spectral signatures with spatial connectivity constraints.

Warning

Agglomerative clustering with spatial connectivity is conceptually elegant, but it doesn’t scale well to large 2D grids, and processing very large datasets can lead to high computing time.

Example

The following example demonstrates spatial agglomerative clustering on a synthetically generated DataCube.

import wizard
from wizard._utils.example import generate_pattern_stack
from wizard._processing.cluster import spatial_agglomerative_clustering, pca
import matplotlib.pyplot as plt

# generate synthetic data with spatial patterns
data = generate_pattern_stack(8, 150, 150, n_circles=2, n_rects=4, n_triangles=3)

# create a DataCube
dc = wizard.DataCube(data)

# visualize a wavelength slice and the segmentation
fig, axes = plt.subplots(1, 2, figsize=(10, 5))

axes[0].imshow(dc[4])
axes[0].set_title("Original Slice (λ index 4)")
axes[0].set_axis_off()

# pca, because agglomerative donst scale well
dc_reduced = pca(dc, n_components=2)

# apply spatially-constrained agglomerative clustering
labels_agg = spatial_agglomerative_clustering(dc_reduced, n_clusters=7)

axes[1].imshow(labels_agg)
axes[1].set_title("Segmented (Agglomerative + Spatial)")
axes[1].set_axis_off()

plt.tight_layout()
plt.show()

The output illustrates how the algorithm merges spectrally similar and spatially adjacent regions, yielding a smooth and interpretable segmentation reflecting both spectral and spatial continuity.