OBIA Classification

Object-Based Image Analysis (OBIA) first segments neighboring pixels into image objects, calculates object attributes, then classifies those objects. This reduces the purely pixel-by-pixel view and can use shape, texture, and context.

Remote Sensing CoreEnd-User DocumentationTheory + Formula + Parameters

1. What This Feature Does

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

2. Recommended Workflow

Select spectral bands. ↓ Segment the image using scale and spectral/shape weights. ↓ Calculate spectral, shape, texture, and context attributes. ↓ Choose Supervised or Unsupervised classification. ↓ Classify objects and review object boundaries/classes.

3. Theory, Methods & Equations

Image Segmentation

Creates spatially connected objects by balancing spectral heterogeneity and shape constraints.

Cost ≈ w_spectral·ΔSpectral + w_shape·ΔShape

Compactness = 4πA/P²

Best used when: meaningful objects are larger than individual pixels.

Random Forest (Supervised)

Classifies object feature vectors with an ensemble of decision trees.

ĉ=mode{f_t(x)}

Best used when: object attributes form nonlinear class boundaries.

SVM (Supervised)

Finds a maximum-margin boundary between object classes, optionally through a nonlinear kernel.

min ½||w||² + CΣξ_i

Best used when: labelled objects are limited but feature separation is strong.

XGBoost / LightGBM (Supervised)

Sequentially adds decision trees that correct residual classification errors using gradient boosting.

F_m(x)=F_m−1(x)+η f_m(x)

Best used when: tabular object features and nonlinear interactions dominate.

MLP Neural Network (Supervised)

Learns nonlinear mappings from object features to class probabilities.

h=φ(Wx+b)

Best used when: enough labelled objects exist for a neural classifier.

K-Means / GMM / BIRCH (Unsupervised)

Groups object-level feature vectors without class labels using centroid, probabilistic, or hierarchical clustering.

K-Means: J=Σ||x−μ||²

GMM: p(x)=Σπ_k𝒩(x|μ_k,Σ_k)

Best used when: exploring object groups before semantic labelling.

4. Input Data

Input	Type	Requirement	Description
Input Raster	raster	Required	Single multiband raster used for both object segmentation and object-feature extraction.
Training Area (ROI)	vector	Required for selected method	Polygon ROI layer used only in Supervised mode. Disabled in Unsupervised mode. Each ROI must contain the selected Class Field.

5. Parameters Available in the Application

Parameter	Default	Choices / Range	How to Use It
Input Bands	—	—	Bands used for segmentation and object features. Leave empty to use All bands.
Scale Parameter	50.0	2.0 to 1000.0	Controls approximate object size. Higher values create larger image objects; lower values create finer segmentation.
Spectral Weight	0.8	0.0 to 1.0	Relative importance of multiband spectral similarity during segmentation. Spectral and Shape weights are normalized internally.
Shape Weight	0.2	0.0 to 1.0	Relative importance of spatial/shape regularity during segmentation.
Compactness	0.5	0.0 to 1.0	Favours compact image objects. Compactness and Smoothness are normalized internally.
Smoothness	0.5	0.0 to 1.0	Favours smoother object boundaries and applies mild pre-segmentation smoothing.
Minimum Object Size (pixels)	16	2 to 100000	Minimum expected object size used by connectivity cleanup.
Maximum Objects	25000	10 to 250000	Safety limit for the number of output OBIA objects.
Spectral Attributes	On	—	Mean, standard deviation, minimum and maximum for every selected band.
Shape Attributes	On	—	Area, perimeter, compactness, elongation, solidity, extent and eccentricity.
Texture Attributes	On	—	Object-level local variance and edge/gradient statistics from the first selected raster band.
Context Attributes	On	—	Neighbour count and mean spectral distance to adjacent objects.
Classification Mode	unsupervised	Supervised, Unsupervised
Supervised Algorithm	—	Random Forest, Support Vector Machine (SVM), XGBoost, LightGBM, MLP Neural Network	Shown for: supervised
Class Field	—	—	ROI attribute containing the target class name/code. Shown for: supervised
Minimum ROI Overlap	0.1	0.0 to 1.0	Minimum fraction of an image object covered by one ROI class before that object is accepted as a supervised training object. Shown for: supervised
Random Forest Trees	300	10 to 5000	Shown for: random_forest
SVM C	10.0	0.0001 to —	Shown for: svm
SVM Gamma	scale	Scale, Auto	Shown for: svm
MLP Hidden Layers	[64, 32]	—	Shown for: mlp
MLP Maximum Iterations	300	10 to 5000	Shown for: mlp
Unsupervised Algorithm	kmeans	K-Means, Gaussian Mixture Model (GMM), BIRCH	Shown for: unsupervised
Number of Classes	5	2 to 100	Shown for: unsupervised
Standardize Object Features	On	—
Random State	42	—

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.

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.

← Unsupervised Classification Image Matching Detection →