Hyperspectral Image Processing Training Course

Participants can organise an ENVI project and inspect imagery using essential display tools.

• ENVI interface panels, layer manager and data manager configured
• Project folders, naming conventions and relative paths established
• Raster metadata, wavelengths and coordinate systems reviewed within ENVI
• Display stretches, linked views and pixel readouts used purposefully
• Regions of interest created for materials, masks and validation samples
• Spectral profiles extracted from pixels, polygons and transects

Participants can distinguish image value types and prepare calibrated radiance data.

• Digital numbers converted to radiance using sensor calibration coefficients
• Radiance, reflectance and apparent reflectance products differentiated correctly
• Gain, offset and solar irradiance metadata checked for consistency
• Bad detector columns and saturated pixels identified visually
• Scan-line artefacts and striping diagnosed from image statistics
• Calibration provenance recorded for later quality assurance

Participants can select and apply an appropriate atmospheric correction approach.

• Atmospheric absorption regions identified using spectral plots and band metadata
• FLAASH parameters configured from acquisition geometry and atmospheric conditions
• QUAC correction applied when atmospheric metadata remains limited
• Water vapour retrieval evaluated using known absorption features
• Aerosol assumptions tested against image-derived reflectance plausibility
• Reflectance outputs compared with reference spectra and scene knowledge

Participants can assess positional quality and align hyperspectral imagery with GIS reference data.

• Coordinate reference systems verified against project basemaps and vectors
• Orthorectification requirements determined from terrain and sensor geometry
• Ground control points selected using stable, identifiable image features
• Root mean square error interpreted for registration quality decisions
• Resampling methods compared for spectral preservation and spatial alignment
• Image subsets and mosaics prepared without unnecessary spectral alteration

Participants can remove unusable bands and reduce noise while preserving diagnostic information.

• Low signal-to-noise bands excluded using sensor documentation and plots
• Water absorption bands masked before spectral analysis procedures
• Minimum Noise Fraction transformation separates coherent signal from noise
• Destriping filters selected according to detector artefact patterns
• Savitzky-Golay smoothing applied without obscuring absorption features
• Noise-reduction decisions documented with before-and-after spectral evidence

Participants can build, assess and use spectral libraries for material identification.

• USGS spectral library records searched by material and wavelength range
• Continuum removal isolates diagnostic absorption feature positions
• Absorption depth, width and asymmetry measured from reflectance spectra
• Image-derived endmembers compared against laboratory reference signatures
• Illumination, moisture and grain-size effects considered during interpretation
• Spectral libraries organised with provenance, units and collection conditions

Participants can reduce spectral dimensionality and create features for analysis.

• Principal Components Analysis summarises correlated spectral information efficiently
• Minimum Noise Fraction components selected using eigenvalue and image inspection
• Band ratios designed around known diagnostic absorption and reflectance features
• Derivative spectra highlight subtle absorption shoulders and inflection points
• Spectral indices calculated for vegetation, minerals and water constituents
• Feature stacks assembled with documented band-selection rationale

Participants can estimate material fractions in mixed pixels using endmember methods.

• Pixel Purity Index locates spectrally extreme candidate endmembers
• N-FINDR extraction identifies endmembers within multispectral feature space
• Convex geometry concepts explain mixed-pixel spectral relationships
• Linear spectral unmixing estimates fractional material abundance values
• Fully constrained unmixing enforces nonnegative fractions and unit sums
• Residual error images reveal missing endmembers or model mismatch

Participants can detect known target materials using spectral matching algorithms.

• Spectral Angle Mapper measures directional similarity between spectral vectors
• Matched filtering maximises known target response against background variability
• Adaptive Cosine Estimator detects targets under variable illumination conditions
• Mixture-Tuned Matched Filtering combines abundance estimates and infeasibility checks
• Detection thresholds calibrated using target samples and false-alarm patterns
• Target maps reviewed against imagery, field records and contextual GIS layers

Participants can design a supervised classification scheme for hyperspectral data.

• Class definitions linked to operational decisions and observable materials
• Training polygons sampled across illumination, condition and spatial variability
• Class separability assessed using Jeffries-Matusita distance and scatterplots
• Support Vector Machine parameters selected for high-dimensional spectral features
• Random Forest classification configured with balanced training samples
• Training datasets partitioned for model fitting and independent validation

Participants can incorporate spatial objects and context into hyperspectral classification.

• Image segmentation groups neighbouring pixels into meaningful spatial objects
• Scale, shape and compactness parameters tuned against target geometry
• Object statistics summarise spectral, texture and contextual attributes
• Morphological profiles distinguish linear infrastructure from natural surfaces
• Neighbourhood features reduce isolated classification noise and speckle
• Vector boundaries incorporated where operational units guide interpretation

Participants can convert raw classification outputs into usable thematic map products.

• Majority filters applied only after reviewing boundary consequences
• Minimum mapping units defined from decision requirements and image resolution
• Class aggregation aligns detailed materials with reporting categories
• Raster-to-vector conversion evaluated for positional and thematic implications
• Symbology communicates confidence, class hierarchy and material distinctions
• GIS export settings preserve projection, attributes and raster value meaning

Participants can quantify thematic accuracy and report uncertainty appropriately.

• Probability sampling designs produce independent, representative validation samples
• Confusion matrices compare mapped classes against reference observations
• Producer's accuracy and user's accuracy interpreted for different decisions
• Overall accuracy and F1 scores reported with class-level results
• Area-adjusted accuracy estimates account for unequal class sampling
• Uncertainty statements separate sensor, model and reference-data limitations

Participants can compare hyperspectral acquisitions to identify meaningful material change.

• Cross-date co-registration verified before pixel-level change analysis
• Radiometric normalisation reduces acquisition-condition differences between dates
• Spectral change vectors identify magnitude and direction of change
• Post-classification comparison produces material transition matrices
• Temporal spectra distinguish seasonal variation from abnormal surface change
• Change thresholds tested against reference observations and operational tolerances

Participants can build reusable ENVI processing chains for repeated image workflows.

• ENVI Modeler nodes connected into ordered processing workflows
• Input parameters exposed for sensor-specific and project-specific settings
• Conditional branches handle missing bands and alternative correction methods
• Batch processing applies standard workflows across multiple image scenes
• Intermediate outputs named systematically for audit and troubleshooting
• Model documentation records assumptions, inputs, outputs and version changes

Participants can combine hyperspectral results with GIS, field and ancillary data.

• Field spectra imported with wavelength units and measurement metadata
• GPS observations matched to imagery using positional uncertainty tolerances
• LiDAR elevation products support terrain and surface interpretation
• Geology, soils and land-cover layers used as contextual evidence
• Field verification plans prioritise uncertain classes and high-consequence locations
• Ancillary data conflicts documented rather than silently overridden

Participants can select fit-for-purpose methods for common hyperspectral applications.

• Mineral alteration mapping uses absorption features near shortwave infrared wavelengths
• Crop stress assessment combines red-edge shifts and water absorption responses
• Coastal water analysis accounts for adjacency effects and shallow-bottom reflectance
• Contamination screening identifies surface materials requiring field confirmation
• Infrastructure inspection separates roofing, pavement and exposed construction materials
• Method selection links detection performance to decision risk and cost

Participants can prepare an auditable technical record for hyperspectral analysis outputs.

• Processing logs capture software versions, parameters and source datasets
• Metadata records document acquisition conditions, corrections and spatial reference
• Quality-control checklists verify outputs before GIS publication
• Map layouts include scale, legend, accuracy statements and data provenance
• Technical reports distinguish observed evidence from interpretive conclusions
• Peer review criteria test reproducibility, plausibility and decision relevance

Participants can build, review and present a complete documented hyperspectral mapping product for a defined decision question.

• Project question framed with target materials, geography and decision users
• Raw imagery assessed for calibration, geometry and atmospheric correction needs
• Processing workflow built from screening, correction and feature-selection methods
• Spectral evidence and classification approach selected for stated target classes
• Accuracy results, uncertainty limits and field checks reviewed critically
• Final maps, workflow documentation and recommendations presented to reviewers