Decision Fusion at Pixel Level of Multi-Band Data for Land Cover Classification-A Review

Area of Science:

• Remote Sensing
• Geospatial Analysis
• Computer Vision

Background:

• Land cover classification relies on spectral signatures from various sensors.
• Existing methods often analyze spectral bands independently, limiting accuracy.
• Integrating diverse data sources offers potential for improved classification reliability.

Purpose of the Study:

• To enhance land cover classification accuracy by fusing decisions from multiple spectral bands.
• To investigate the benefits of incorporating pixel correlations and neighborhood information.
• To present the mathematical framework for pixel-based classification and decision fusion.

Main Methods:

• Decision fusion of optical, thermal infrared, and PolSAR data.
• Pixel-based classification incorporating inter-pixel correlations.
• Utilizing neighborhood information to refine classifications.
• Analysis of hyperspectral, multispectral, thermal infrared, and SAR data.

Main Results:

• Fusing decisions from different spectral bands significantly improves land cover classification reliability.
• Incorporating pixel correlations and neighborhood context enhances classification accuracy.
• Each data source (hyperspectral, multispectral, thermal, SAR) provides unique, complementary information.

Conclusions:

• Multi-sensor data fusion is a powerful approach for accurate land cover mapping.
• Decision fusion methods, considering spatial context, are crucial for robust classification.
• The presented mathematical framework supports advanced pixel-based classification techniques.