Unmixing Models
Overview
The unmixing model shows results for each endmember in a separate heatmap. The yellow depicts the abundance of the respective material (endmember), so the brighter the yellow, the greater the abundance of that material per pixel.
Use Case
Provide images in the form of heatmaps of abundance for each endmember (i.e. one image per endmember). This can be used to understand if there are contaminants within a material, or if there are multiple materials present within a given pixel.
Example
Confidence Thresholds
To filter out noise, you can toggle on "Confidence thresholds" in the toolbar. Confidence thresholds are set individually for each endmember, and tell the model the minimum percentage of a target endmember per pixel to visually depict.
For example, a confidence threshold of 0 means that the model will identify any pixel where even the faintest detection of a spectral signature of your endmembers was detected. A confidence threshold of 1 means that the model will only identify pixels for which the composition consists exclusively (100%) of the spectral signature of your endmembers.
Depending on the endmembers you are looking to identify, you may adjust the confidence threshold per unmixing result image. Below is an example of fine tuning our model for tree identification using confidence thresholds.
How It Works
Unmixing models excel in accurately determining the composition and proportion of multiple materials within a single pixel, leveraging the rich spectral information in HSI for detailed material analysis. They are especially suited for detailed material analysis and complex scenarios where pixels contain mixtures of several constituents.
Strengths
- Excel in accurately determining the composition and proportion of multiple materials within a single pixel
- Leverage the rich spectral information in HSI for detailed material analysis
- Especially suited for detailed material analysis and complex scenarios where pixels contain mixtures of several constituents