VibraVision
VibraVision is a software platform for spectral data analysis, chemometric modelling, prediction, and instrument performance evaluation.
It brings the complete workflow, from data loading and spectral preprocessing to model development, validation, prediction and monitoring, into one integrated environment.
VibraVision includes the following modules:
1. Model Creation and Validation
A complete workflow for developing, optimizing, and validating regression and classification models.
1.1. Data Set Management
- Load X data, Y data, or combined tables.
- Combine data sets at a later stage.
- Manage spectral and reference data within the same workflow.
1.2. Spectral Visualization
- Interactive visualization of spectra.
- Inspect individual spectra by moving the cursor over the graph.
- Switch between absorbance and transmittance.
- Visualize intensity versus mean intensity.
- Preview the first 10 columns of data tables.
1.3. Spectral Preprocessing
- Apply spectral preprocessing methods according to the selected workflow.
- Compare and evaluate different preprocessing options.
1.4. Outlier Detection and Data Cleaning
- PCA
- UMAP
- DBSCAN
- Isolation Forest
- Mahalanobis distance
Visualize potential outliers, select samples manually, remove unwanted samples, and generate clean data tables.
The cleaned data can then be visualized again to assess the effect of outlier removal.
1.5. Variable Selection
- Select relevant spectral variables.
- Optimize variable-selection algorithms according to the selected workflow.
1.6. Data Set Division
- Divide data into calibration and validation sets.
- Maintain the selected class proportions for classification models, for example 70/30.
- Keep spectra from the same physical sample together so that they cannot be split between calibration and validation.
- Use different data-splitting algorithms, including Kennard-Stone, Random, Duplex, and Onion.
- Manually define calibration and validation samples using specified ranges.
1.7. Model Training and Optimization
- Perform cross-validation within the calibration set.
- Alternatively, divide the calibration data into Calibration and Tuning sets.
- Select the optimal number of latent variables for PLS and PLS-DA.
- Visualize RMSEP as a function of the number of latent variables.
- Compare calibration and tuning performance, or cross-validation performance.
- Identify the point where prediction error reaches its minimum before increasing again.
- Optimize model hyperparameters by allowing users to define the values to be tested.
1.8. Model Validation
Evaluate the calibrated model using an independent validation set.
For regression:
- RMSEC
- RMSEP
- R² Calibration
- R² Prediction
For classification:
- Confusion matrix
- Sensitivity
- Specificity
- Balanced accuracy
- F1 score
2. Prediction & Monitoring
Use an existing calibrated model to predict new samples.
VibraVision allows users to apply a selected model to new spectral measurements for routine prediction and monitoring.
3. Accuracy Profile
Evaluate model performance across the relevant concentration or measurement range and assess the accuracy of predictions.
4. CNN & ANN
A dedicated environment for developing and training Artificial Neural Network and Convolutional Neural Network models using different architectures.
5. Hyperspectral Imaging
Extract and analyze spectral information from hyperspectral images.
Select regions or pixels within images and extract their corresponding spectral data for further analysis and modelling.
6. Calibration Transfer
Apply calibration-transfer algorithms to transfer models between different instruments and maintain model performance across equipment.
7. Instrument Verification
A dedicated module for verifying the correct operation and performance of spectral instruments using nine instrument performance tests.
Want to try VibraVision?
Start with a free trial and explore what VibraVision can do for your spectral data.