Multi-Condition Classification of Oil Spill in Ice Areas Based on Laser-Induced Fluorescence

Area of Science:

• Environmental Science
• Spectroscopy
• Machine Learning

Background:

• Oil spill detection in icy marine environments is challenging due to signal interference and complex conditions.
• Existing methods lack high precision for oil classification under these circumstances.

Purpose of the Study:

• To develop a robust, fluorescence-based framework for multi-condition oil classification in ice-covered waters.
• To enhance the accuracy of oil spill detection and identification in challenging polar environments.

Main Methods:

• Collected laser-induced fluorescence (LIF) spectra for six oil types across four simulated ice/volume scenarios.
• Applied Savitzky-Golay (SG) filtering for fluorescence signal denoising and stability.
• Developed a machine learning model optimized by the Golden Sine Algorithm (Gold-SA) for classification.

Main Results:

• The Gold-SA-CatBoost model achieved 99.62% accuracy on the dataset and 100% for single-task oil identification.
• The framework demonstrated superior performance compared to baseline models.
• Successfully classified 24 distinct categories under various ice coverage and oil volume conditions.

Conclusions:

• The integration of LIF spectroscopy with Gold-SA optimized machine learning offers a viable strategy for oil spill detection in icy conditions.
• This approach significantly enhances environmental monitoring capabilities in cold and polar marine regions.
• The study addresses the need for high-precision oil classification in complex, ice-affected marine ecosystems.