Summary of Model Types

Overview

Clarity offers four main types of machine learning models, each designed for specific analysis tasks. Understanding the differences between these model types will help you choose the right approach for your hyperspectral data analysis needs.

Model Type Comparison

Model Type	Purpose	Output	Best For	Example Use Cases
Classification	Categorize pixels into predefined classes	Color-coded class map	Land cover mapping, material identification	Vegetation types, land use classification
Unmixing	Determine material abundance within pixels	Abundance heatmaps	Mixed pixel analysis, material composition	Mineral mapping, contaminant detection
Target Detection	Find specific materials or objects	Probability maps	Rare target identification	Vehicle detection, mineral prospecting
Regression	Predict continuous values	Continuous value maps	Quantitative measurements	Chlorophyll content, soil moisture

Classification Models

What they do: Assign each pixel to a specific class or category based on spectral signatures.

Key characteristics:

• ✅ Simple to understand - Each pixel gets one class label
• ✅ Great for mapping - Creates clear categorical maps
• ✅ Wide applicability - Works for most land cover and material identification tasks
• ⚠️ Requires distinct classes - Classes must have different spectral signatures

Best when: You need to categorize materials, land cover types, or create thematic maps.

Learn more about Classification →

Unmixing Models

What they do: Determine the proportion of different materials present within each pixel.

Key characteristics:

• ✅ Handles mixed pixels - Perfect for complex, mixed-material scenarios
• ✅ Quantitative results - Provides abundance percentages for each material
• ✅ Detailed analysis - Shows material composition at sub-pixel level
• ⚠️ More complex interpretation - Results require understanding of abundance values

Best when: You're analyzing pixels that contain mixtures of materials or need to understand material proportions.

Learn more about Unmixing →

Target Detection Models

What they do: Identify specific materials or objects of interest within an image.

Key characteristics:

• ✅ Focused detection - Designed to find specific targets with high confidence
• ✅ Efficient processing - Optimized for finding rare or specific materials
• ✅ High precision - Minimizes false positives for target materials
• ⚠️ Limited scope - Only identifies the specific target(s) you train for

Best when: You're looking for specific materials, objects, or need to detect rare occurrences.

Learn more about Target Detection →

Regression Models

What they do: Predict continuous numerical values for each pixel based on spectral data.

Key characteristics:

• ✅ Quantitative predictions - Provides actual numerical values, not just categories
• ✅ Continuous mapping - Shows gradual changes across the landscape
• ✅ Physical measurements - Can predict chemical concentrations, indices, etc.
• ⚠️ Requires labeled attributes - Training data needs numerical values, not just class labels

Best when: You need to measure continuous variables like chemical concentrations, vegetation indices, or environmental parameters.

Learn more about Regression →

Choosing the Right Model Type

Start with Your Goal

1. What do you want to achieve?

   • Map different land cover types? → Classification
   • Understand material mixtures? → Unmixing
   • Find specific targets? → Target Detection
   • Measure continuous values? → Regression

2. What does your data look like?

   • Clear, distinct classes? → Classification
   • Mixed materials in pixels? → Unmixing
   • Specific target signatures? → Target Detection
   • Continuous variables to predict? → Regression

3. What output do you need?

   • Categorical maps? → Classification
   • Abundance maps? → Unmixing
   • Target probability maps? → Target Detection
   • Continuous value maps? → Regression

Common Combinations

• Land cover analysis: Start with Classification, then use Unmixing for detailed material analysis
• Mineral exploration: Use Target Detection for specific minerals, then Classification for broader mapping
• Environmental monitoring: Combine Regression for continuous measurements with Classification for land cover

Getting Started

Ready to train your first model? Start with the Model Training guide to learn the step-by-step process.

Need help choosing? Our team can guide you through model selection based on your specific use case. Contact us at support@metaspectral.com for personalized assistance.