Dashboard
Viewing classification results, statistics, and quality metrics
The Dashboard screen provides a comprehensive visualization of classification results after AI analysis. It displays class distributions, screen size breakdowns, scatter plots, confusion matrices, weight estimation, and density analysis. Multiple view modes and filters let you explore the data from different perspectives.
Prerequisites
An analysis file with completed AI classification results.
Screen Layout
The screen uses a tabbed interface with multiple display modes:
Dashboard: Main results overview with class distribution and statistics.
Method-wise Results: Classification equivalence comparisons across methods.
Within the Dashboard tab, additional view modes are available:
All Images
Shows results for all classified seeds
By Class
Groups results by classification category
By Screen Size
Breaks down results by sieve size
Step-by-Step Walkthrough
1. Open the Dashboard
Navigate to the Dashboard screen from the sidebar after running AI classification. The application automatically loads the most recent analysis results from the current HDF5 file.
2. Review the Class Distribution
The main panel displays a bar chart showing the number of seeds in each classification category (e.g., OK, Black, Sour, Broken). The chart uses color-coded bars matching the model's defined hue colormap.
Key metrics displayed alongside the chart:
Total Seeds: Number of seeds analyzed.
OK Percentage: Proportion of seeds classified as acceptable.
Defect Breakdown: Count and percentage for each defect category.
3. Filter by Screen Size
Use the screen size range filter to narrow results to a specific sieve range. This is useful when analyzing grade-specific quality (e.g., only seeds that passed through sieve 14 but were retained by sieve 15).
4. Explore Scatter Plots
The scatter plot panel lets you visualize relationships between any two extracted features:
X-axis / Y-axis: Select from available features (area, perimeter, color channels, entropy, circularity, etc.).
Colormap: Choose how points are colored (by class, by screen size, by feature value).
Each point represents a single seed. Hover to see detailed feature values.
Scatter plots are useful for identifying clusters and outliers. Try plotting area vs. entropy to spot seeds that the model may have had difficulty classifying.
5. Review Weight Estimation
The weight estimation section shows:
Estimated weight per class based on 2D area measurements and density maps.
Relative to count vs. relative to weight toggle for percentage calculations.
Density parameters from the loaded AI model.
You can update seed weight values directly from the dashboard if corrections are needed.
6. View Confusion Matrix
If the analysis includes reference classifications, the confusion matrix shows agreement between the AI model's predictions and the reference labels, with precision and recall metrics per class.
7. Filter Moka (Peaberry) Subset
If the analysis contains Moka beans, use the Moka filter to:
View only Moka beans.
View only regular beans.
View all beans combined.
The Moka filter is only visible when the analysis file contains Peaberry bean data and the "Display Moka" toggle is enabled in General Settings.
8. Save Visualizations
Click the save icon on any chart to export it as a PNG image. A file dialog appears to choose the save location.
Entropy Analysis
Click the entropy help icon to open the Entropy Modal, which explains:
How entropy measures classification uncertainty.
What high vs. low entropy values indicate.
How to interpret entropy distributions across your sample.
Configuration Persistence
Dashboard preferences (relative switch, Moka display, selected view mode) are persisted in the Electron Store and restored on subsequent visits.
Troubleshooting
Dashboard shows no data
No AI analysis has been run on this file
Run AI Classification first
Charts are empty
Analysis file was opened but results were cleared
Re-run AI Classification
Weight values seem incorrect
Density map not configured in the model
Edit density parameters in Edit Model
Moka filter not visible
No Moka data in file or display toggle is off
Check General Settings for "Display Moka" toggle
Screen size filter has no effect
All seeds are the same sieve size
This is expected when the sample has uniform screen size
Last updated