Multispectral vision

Turn spectral images into decisions

When color and shape are insufficient, additional wavelengths can reveal useful differences. Abraia helps teams inspect spectral data, label examples, develop models, and test application-specific workflows.

Choose the task

From spectra to a testable result

The right method depends on what must be predicted and how the data was annotated.

Object classification and regression

Predict a class or numeric target for an existing segmented object from its spectral data.

Explore object models →

RGB plus spectral pipelines

Locate objects in an RGB view, then classify their spectral crops in a later stage.

Explore the pipeline →

01 · Inspect and prepare data

Understand the bands before training

Open supported TIFF, ENVI, or IMEC scenes in Vision Studio. Compare true-color, false-color, single-band, PCA, and custom views; click pixels to inspect their spectra and review source metadata.

Spectral indices depend on wavelength metadata and the bands available in the source. A useful analysis checks acquisition conditions and whether the signal separates the classes of interest. A peppers scene demonstrates annotated object analysis across 31 bands.

Vision Studio false-color view of peppers with a selected band and spectral signature
False-color multispectral view · Inspect bands, signatures, and source metadata.

02 · Choose and test a model

Build a workflow from spectra to predictions

Use labeled areas for pixel classification and inspect a predicted mask. For segmented objects, train a classifier or regressor to predict a class or numeric target from spectral data.

For an RGB-plus-spectral pipeline, locate objects in the RGB view, then classify their spectral crops in a later stage. The Studio example processes strawberries with instance segmentation followed by an ONNX spectral classifier. These examples show the workflow on specific datasets; performance on new data needs project-specific evaluation.

Vision Studio strawberry pipeline with instance segmentation and ONNX spectral classification
One sample frame produced seven detections; this is a workflow example, not a benchmark.

03 · Project evaluation

Evaluate spectra for the target material

Spectral methods are worth evaluating when materials or conditions look similar in ordinary color images. Potential uses include produce inspection, plant monitoring, material sorting, and quality assessment. Feasibility depends on the sensor, wavelength range, illumination, and reference data.

For crop and produce examples, see computer vision in agriculture. For local camera systems, see embedded vision solutions.

Vision Studio false-color view of annotated peppers with spectral model controls
Annotated 31-band peppers scene.

Bring your vision problem to us

Tell us about the images, sensor, operating environment, and decision you want the system to support. We can identify a practical first evaluation.