Principal Component Analysis

1. Overview

Principal Component Analysis (PCA) is an unsupervised dimensionality-reduction technique that transforms the original spectral bands into a new set of uncorrelated components - ordered from highest to lowest explained variance. In IDCubeCloud, PCA works on any loaded hyperspectral or satellite cube and is the recommended first step before clustering or spectral analysis.

Key points

• No training data required - PCA is entirely data-driven.
• Reduces noise by concentrating most variance into the first few components.
• Output components are fully compatible with all other IDCubeCloud tools (Clustering, 3D Cube, export).

2. What PCA Does

1. Original hyperspectral cube [H x W x B]

2. Each pixel is treated as a point in B-dimensional spectral space

3. PCA computes the covariance matrix of all pixel spectra

4. Eigendecomposition to principal components (eigenvectors) sorted by explained variance (eigenvalues, largest first)

5. Pixels are projected onto the top N components

6. Output: N component images [H x W] + explained-variance report

Why this matters: Satellite and hyperspectral data have highly correlated adjacent bands. PCA decorrelates them, separating meaningful spectral variation (land cover, vegetation, moisture) from sensor noise.

3. Where to Find It

1. Load any data file from My Files in the sidebar.

2. The toolbar at the top of the workspace will show: Explore | Measure | PCA | Filters | Spectral Library | Indices | Clustering | Machine Learning, plus a separate 3D View button.

3. Click PCA in the toolbar to the PCA Parameters panel appears in the left sidebar.

4. Step-by-Step Workflow

Step 1 Load Data

Load a .mat, ENVI, TIFF, HDF5, NIfTI, AVI, NPY, or NetCDF file - or load a satellite scene from the Satellite page.

Step 2 Open PCA Panel

Click PCA in the feature toolbar. The sidebar shows the PCA Parameters section.

Step 3 Set Number of Components

Set the Components spinner to the number of principal components to compute.

Step 4 Run PCA

Click Compute PCA. Processing time scales with image size. A loading indicator shows while the server computes the decomposition.

Step 5 Inspect Results

• Component images appear in the main viewer - use the band slider to scroll through PC1, PC2, etc.
• Loading coefficients chart shows which original bands contribute most to each component.
• Variance report lists the % explained variance per component.

5. Parameters Explained

6. Output and Interpretation

Interpreting the variance report

• If PC1 explains > 90% of variance, the data is highly uniform (single land-cover type, low contrast scene).
• If the first 3-5 components explain > 95% collectively, those components capture most meaningful information and can be safely used for clustering.

7. When to Use PCA

8. Known Limitations

9. Common Use Cases

Part of the IDCubeCloud User Documentation - see also: Clustering Guide, Spectral Library Guide, Vegetation Indices Guide.

Manual revision: September 2026