Unsupervised Classification

Automatically groups pixels with similar multiband values without labelled training data. The result is a cluster map that must be interpreted before it becomes a semantic land-cover map.

Remote Sensing CoreEnd-User DocumentationTheory + Formula + Parameters

1. What This Feature Does

Automatically groups pixels with similar multiband values without labelled training data. The result is a cluster map that must be interpreted before it becomes a semantic land-cover map.

Cloud workflow: choose data → configure the scientific method → run → review the resulting layer/report. No programming is required.

2. Recommended Workflow

Select informative predictor bands. ↓ Keep standardization on when numeric scales differ. ↓ Choose a clustering algorithm. ↓ Set the class count or adaptive bounds. ↓ Run the clustering. ↓ Interpret and, if necessary, merge/rename spectral clusters.

3. Theory, Methods & Equations

K-Means

Minimizes within-cluster squared distance to K centroids.

J=Σ||x_i−μ_c(i)||²

μ_k=(1/|C_k|)Σx_i

Best used when: a fast general baseline with a fixed class count is desired.

ISODATA

Adaptive centroid clustering that can remove small clusters, merge nearby centers, and split high-variance clusters.

Merge if ||μ_i−μ_j|| < T_merge

Split when max(σ_k,b) ≥ T_split

Best used when: the natural number of clusters is uncertain.

Self-Organizing Map (SOM)

Learns a 2-D topology of representative feature vectors, then groups map units into output classes.

BMU(x)=argmin_j||x−w_j||²

h_bj=exp(−d_grid²/(2σ²))

Best used when: complex spectral topology is worth exploring.

Gaussian Mixture Model (GMM)

Models data as a mixture of Gaussian components with configurable covariance.

p(x)=Σπ_k𝒩(x|μ_k,Σ_k)

Best used when: clusters overlap or have elliptical covariance structure.

Fuzzy C-Means

Assigns soft membership to all clusters during fitting, then produces a class map from the strongest membership.

u_ik=1/Σ_j(d_ik/d_ij)^2/(m−1)

c_k=Σu_ik^mx_i/Σu_ik^m

Best used when: mixed pixels and gradual spectral transitions are common.

BIRCH

Builds compact Clustering Feature summaries in a hierarchy, then performs global grouping.

CF=(N,LS,SS)

μ=LS/N

Best used when: efficient clustering of large feature samples is needed.

4. Input Data

Input	Type	Requirement	Description
Input Raster	raster	Required

5. Parameters Available in the Application

Parameter	Default	Choices / Range	How to Use It
Classification Algorithm	—	K-Means, ISODATA, Self-Organizing Map (SOM), Gaussian Mixture Model (GMM), Fuzzy C-Means, BIRCH	Choose the clustering method.
Predictor Bands	All bands	—	Select raster bands used to form each pixel's feature vector.
Number of Classes	5	2 to 255	Requested classes; for ISODATA this is the initial class count.
Training Pixel Sample	100000	100 to —	Maximum valid pixels sampled to learn the cluster structure.
Standardize Predictor Bands	On	—	Recommended when bands or stacked variables have different numeric ranges.
Prediction Batch Size	250000	1000 to 2000000	Controls memory used while assigning the final raster classes.
Random State	42	—	Keep fixed to make comparative runs more reproducible.
Output Nodata	0	0 to 32767	Value reserved for invalid / nodata output pixels.
K-Means Mini-Batch Size	4096	64 to 262144	Shown for: kmeans
K-Means Maximum Iterations	100	1 to 2000	Shown for: kmeans
K-Means Initializations	3	1 to 50	Shown for: kmeans
ISODATA Minimum Classes	2	2 to 255	Shown for: isodata
ISODATA Maximum Classes	12	2 to 255	Shown for: isodata
ISODATA Maximum Iterations	30	1 to 500	Shown for: isodata
ISODATA Minimum Pixels per Cluster	50	1 to 1000000	Shown for: isodata
ISODATA Merge Distance Threshold	0.5	0 to —	Small distances encourage merging of similar clusters. Shown for: isodata
ISODATA Split Standard Deviation Threshold	1.0	0 to —	High within-cluster variation can trigger splitting. Shown for: isodata
ISODATA Convergence Tolerance	0.01	1e-07 to —	Shown for: isodata
SOM Grid Rows	5	1 to 64	Shown for: som
SOM Grid Columns	5	1 to 64	Shown for: som
SOM Epochs	30	1 to 1000	Shown for: som
SOM Initial Learning Rate	0.5	1e-06 to 2.0	Shown for: som
SOM Initial Neighborhood Sigma	2.0	0.1 to 100	Shown for: som
SOM Mini-Batch Size	1024	16 to 65536	Shown for: som
GMM Covariance Type	Full	Full, Tied, Diagonal, Spherical	Shown for: gmm
GMM Maximum Iterations	200	1 to 5000	Shown for: gmm
GMM Convergence Tolerance	0.001	1e-08 to —	Shown for: gmm
GMM Covariance Regularization	1e-06	0 to —	Shown for: gmm
GMM Initializations	1	1 to 50	Shown for: gmm
Fuzzy C-Means Fuzziness (m)	2.0	1.0001 to 10	Shown for: fuzzy_cmeans
Fuzzy C-Means Maximum Iterations	150	1 to 5000	Shown for: fuzzy_cmeans
Fuzzy C-Means Convergence Tolerance	0.0001	1e-08 to —	Shown for: fuzzy_cmeans
Fuzzy C-Means Initializations	1	1 to 20	Shown for: fuzzy_cmeans
BIRCH Threshold	0.5	1e-06 to —	Shown for: birch
BIRCH Branching Factor	50	2 to 10000	Shown for: birch

6. Output & Interpretation

The result should be interpreted according to the selected method and the physical meaning of the input data. Preserve source units, coordinate reference information, NoData meaning, acquisition date, and preprocessing level when comparing results.

Scientific interpretation: an algorithm can produce a numerically valid result even when the input data are unsuitable. Always check masks, units, sensor characteristics, spatial resolution and reference data.

Important: Cluster IDs are arbitrary. Class 1 in one run does not necessarily represent the same surface type in another run.

7. Best Practices

Use analysis-ready inputs and remove invalid/cloud/noise artifacts that are not part of the target phenomenon.
Choose parameters from the physical scale of the data, not only from visual appearance.
Keep categorical and continuous rasters conceptually separate when selecting interpolation/resampling methods.
Compare the result with the source image and independent reference information.
Document the settings used when results will be compared across dates, sensors or study areas.

← Supervised Classification OBIA Classification →