v2022.1 / v2022.1.5 (March 1, 2022)

01 Mar 2022

^{Feature walk-through: Improved interaction between your classification/quantification models and the image data}

Modeling and data analysis

• Improved the interaction between models and image data, making easier to add new training data, retrain models and view the results on your image

  • Apply models on images directly in the Record Study table

  • Include, exclude or balance new training samples and then retrain models

  • Update categories and properties data from model prediction results

• ONNX models inference performance improvements

• Constrained spectral unmixing with new options for Scatter correction parameter "None" and method for "Sum To One Least Squares" Unmixing (descriptor) Unmixing (segmentation)

• Better naming for the different machine learning algorithms

• More statistics in the classification confusion matrix (f-score, precision and recall)

• Prediction summary (Observed vs Predicted) for quantification analysis through right-click in Table

• Pre-treatment option to use SNV in the Explore tab

• Use multiple spectral ranges for SAM (Spectral Angle Mapper) Spectral angle mapper (SAM) guide

• Balance model with included train data or all train data

Image segmentation

• Deep Learning Image Segmentation (Faster R-CNN and Yolo4/5) applied to pseudo RGB or model predicted image ONNX image segmentation

• Image segmentation using YOLOv5

• K-Means clustering as method for selecting representative spectrum Representative spectrum

• Add samples to Manual segmentation through right-click in Table

• Object filter for segmentation (ex. object shape)

• Summary table showing pixel overlap of multiple segmentations

• Setting for Max number of objects in Deep Learning Image Segmentation

Visualization

• Pixels on an object that are excluded in the segmentation (e.g. holes showing the background) are also excluded in the visualizations (Pixel Explore, Table thumbnail)

• Merge all samples in segmentation now show all merged samples as one object in the preview image and in Table

• Change RGB bands by right clicking on image

• Test scan show % saturated bands

• Selected samples in Pixel Explore can be highlighted in Table and vice versa

• Option for Quantification model to predict and show value for individual pixels or only object average

• Change direction of real time visualization (horizontal or vertical)

• Option to show all or individual segmentation in Table if there are multiple segmentations on the same level

• Select multiple samples in preview image by holding down the shift key and making rectangular selection of the area with the mouse

• Yellow color for added class in Classification summary (confusion matrix)

Workflow 

• Disable nodes in Analyse Tree that should not be used

• Export and import Breeze models into your study

• White and dark reference descriptor (signal intensity value)

• Metadata descriptor for each image (ex. image length and resolution)

• Delete workflow command in Breeze Runtime

• Access Study settings in Analyse Tree on Measurement node

• Link multiple segmentations to one output

Data acquisition

• Stray light correction option in Settings 

• Saturation descriptor for maximum value and percent saturated band per pixel

• Explore pixel spectra on images in Test Scan

• Indication that new white or dark reference will be taken when recording using reference cache

• Set integration time in Record wizard

General ease-of-use improvements

• Duplicate Study with option to include all measurements and data

• Move measurement to another Group in the same Study

• Improved help sections for Segmentations , Descriptors and Actions

• New tutorial and data included: Classification of plastics Plastic Classification