Skip to main content

Layer Correction Tool

Tweak and correct segmented layers to ensure your training data is clean and reliable

Overview​

The Layer Correction Tool enhances classification accuracy by refining class boundaries using spectral data from imagery bands or vegetation indices. This is especially useful for improving segmentation between similar features where mixed classifications occur.

When to Use Each Layer Correction Option​

Layer corrections in TytonAI can be performed using individual bands or vegetation indices. Below are recommendations and use cases for each:

🔴 RED Band​

Best for: Detecting bare ground or light/dark contrast.
Use case: Isolating exposed soil or dry earth from vegetation.
Why: The RED band shows strong reflectance differences between vegetated and non-vegetated surfaces.

➡️ Example: Ground is being misclassified as tussock — use RED to clean up ground areas.

🟢 GREEN Band​

Best for: Differentiating vegetation types.
Use case: Separating light green vs. dark green plants or fresh vs. mature growth.
Why: GREEN band captures variations in chlorophyll concentration.

➡️ Example: Distinguishing young shrubs from mature ones in mixed terrain.

🔵 BLUE Band​

Best for: Identifying water, shadows, or cool-toned areas.
Use case: Classifying wet soil, shallow water, or shaded canopy regions.
Why: BLUE is sensitive to atmospheric scattering and water reflectance.

➡️ Example: Identifying seasonal water bodies or shaded regions in high resolution drone imagery.

🌿 CIVE (Colour Index of Vegetation Extraction)​

Best for: Vegetation detection in RGB-only imagery.
Use case: Extracting vegetation without near-infrared (NIR) data.
Why: CIVE is optimized for true-colour datasets.

➡️ Example: Working with drone imagery that lacks multispectral bands, needing to identify plant cover.

🌱 MSAVI (Modified Soil-Adjusted Vegetation Index)​

Best for: Areas with sparse vegetation and exposed soil.
Use case: Enhancing detection of plant cover in arid, degraded, or mixed-soil environments.
Why: MSAVI reduces the impact of soil brightness.

➡️ Example: Classifying tussock patches in dry rangelands with lots of bare ground.

🟩 VGI (Vegetation Greenness Index)​

Best for: Highlighting healthy, vigorous vegetation.
Use case: Differentiating lush green cover from stressed or less vigorous plants.
Why: VGI emphasizes greenness and vitality.

➡️ Example: Isolating healthy shrubs in a restoration zone or dense plant growth for biomass estimation.

Step-by-Step Workflow​

1. Open the Layer Correction Tool​

Click on the Layer Correction icon in the toolbar.

This opens a split-screen comparison:

  • Left side: Master Layer (e.g. your training area.)
  • Right side: Adjustment Layer (e.g. RED band, NDVI, etc.)

2. Select Layers​

From the dropdown menus:

  • Master Layer: Select the dataset you want to update (e.g. Training Area).
  • Adjustment Layer: Choose a band or index, such as:
    • RED – For reflectance contrast and highlighting exposed soil or stressed vegetation.
    • GREEN – Useful for identifying healthy vegetation and chlorophyll response.
    • BLUE – Often highlights water bodies and can assist in atmospheric correction.
    • MSAVI – Modified Soil-Adjusted Vegetation Index; reduces soil background influence in sparse vegetation.
    • VGI – Vegetation Greenness Index; emphasizes green canopy presence.
    • NDVI – Normalized Difference Vegetation Index; a common index for vegetation health and density.
    • CIVE – Color Index of Vegetation Extraction; helpful in distinguishing vegetation from non-vegetation in RGB imagery.

3. Analyse the Histogram​

The right panel displays a histogram for the selected adjustment layer.

  • Low values (e.g., 0–80): Dark areas (typically ground).
  • High values (e.g., 180–255): Light/reflective areas (e.g. vegetation).

Set a threshold range (e.g. 178–255) to isolate specific features.

4. Define Correction Rules​

Use the Correction Rules section to:

  • Specify conditions: e.g. Where RED > 178
  • Assign new class: e.g. set to Ground
  • Add filters: Limit updates to areas currently unclassified or already labelled, for example Tussock

This ensures you're not overwriting unrelated features like shrubs or trees.

5. Preview and Validate​

Adjust transparency to compare before/after results side-by-side.

Use this view to:

  • Validate accuracy.
  • Confirm boundaries.
  • Detect noise or over-segmentation.

6. Apply the Correction​

Once satisfied, click "Burn in to Master". This action:

  • Updates your original training dataset.
  • Adds a new version (e.g. Training Area v2).
  • Displays new polygon stats (e.g. 17,000 ground features added).

⚠️ Note: High polygon count may affect performance. Consider filtering or removing very small features via the Feature Table.

Optimization Tips​

  • Open the Feature Table to remove small, noisy polygons.
  • Filter by area to find and delete polygons under a certain threshold. (e.g. Area<=0.001m2).
  • Use correction rules to exclude specific classes from being affected (e.g. Shrub, Tree).

Key Benefits​

  • Significantly improves classification boundaries
  • Supports rule-based updates using spectral logic
  • Enhances model performance and training data quality
Tip
  • Begin with a narrow threshold range and expand gradually to avoid over-segmentation.
  • When applying correction rules, always exclude existing well-classified classes (like Shrub or Tree) to avoid unintentionally altering them.
  • Use the Feature Table to identify and delete very small or noisy polygons that may result from the correction.
  • Use descriptive names for new training versions (e.g. Training Area v2 – Red Band Ground Fix) to maintain clarity in your workflow history.