Sedimentary structure discrimination with hyperspectral imaging in sediment cores

Area of Science:

• Geology
• Remote Sensing
• Machine Learning

Background:

• Sedimentary structures in lake cores record geodynamical events.
• Classical methods for sediment analysis are slow and lack spatial resolution.
• Accurate identification of sediment layers is crucial for hazard assessment.

Purpose of the Study:

• To compare supervised classification algorithms for discriminating sedimentological structures in lake sediments.
• To evaluate the effectiveness of Visible Near-Infrared (VNIR) and Short Wave Infrared (SWIR) hyperspectral imaging sensors.
• To improve the characterization of sedimentary structures using advanced imaging and machine learning.

Main Methods:

• Applied VNIR (400-1000 nm) and SWIR (1000-2500 nm) hyperspectral imaging sensors to three lake sediment cores.
• Utilized supervised classification algorithms and discriminant analyses for sediment layer discrimination.
• Compared spatial and spectral pre-processing techniques for optimal data analysis.

Main Results:

• The SWIR sensor achieved high prediction accuracies (0.87-0.98) for robust classification models.
• VNIR sensor performance was compromised by surface variations, leading to mis-classifications.
• Optimized pre-processing highlighted sensor-specific discriminant information.

Conclusions:

• Hyperspectral imaging, particularly with the SWIR sensor, significantly enhances the characterization of sedimentary structures.
• Machine learning integration with HSI provides a more accurate and efficient alternative to conventional sediment analysis methods.
• This approach improves the reconstruction of past geodynamical events and hazard assessments.