Zhao Chenyu, Li Ying, Xu Qintuan, Wang Yong, Xie Ming, Ji Xiangxiang
Navigation College, Dalian Maritime University, 1 Linghai Road, Dalian, 116026, China.
J Fluoresc. 2025 Sep 16. doi: 10.1007/s10895-025-04547-w.
Oil spill detection in ice-covered marine environments poses considerable challenges due to fluorescence signal interference from ice, heterogeneous surface properties, and environmental complexity. To address the lack of high-precision oil classification methods under such conditions, this study introduces a fluorescence-based multi-condition classification framework that integrates laser-induced fluorescence (LIF) spectroscopy with a machine learning model optimized by the Golden Sine Algorithm (Gold-SA). LIF spectra were collected for six oil types under four simulated ice coverage and oil volume scenarios, resulting in 24 distinct classification categories. Fluorescence signals underwent denoising using Savitzky-Golay (SG) filtering to improve signal stability and spectral reliability. The resulting Gold-SA-CatBoost model achieved 99.62% accuracy under laboratory conditions within the dataset and 100% accuracy in single-task oil-type identification, surpassing baseline models by a substantial margin. This work demonstrates the efficacy of integrating LIF with advanced optimization-based machine learning for robust oil spill detection under complex icy conditions. The proposed approach provides a viable fluorescence-based strategy for environmental monitoring in cold and polar marine regions.
在冰封的海洋环境中进行溢油检测面临着诸多挑战，这是由于冰的荧光信号干扰、表面性质不均以及环境复杂所致。为解决在这种条件下缺乏高精度油品分类方法的问题，本研究引入了一种基于荧光的多条件分类框架，该框架将激光诱导荧光（LIF）光谱与通过黄金正弦算法（Gold-SA）优化的机器学习模型相结合。在四种模拟的冰覆盖率和油量场景下，收集了六种油类的LIF光谱，从而产生了24种不同的分类类别。使用Savitzky-Golay（SG）滤波对荧光信号进行去噪，以提高信号稳定性和光谱可靠性。所得的Gold-SA-CatBoost模型在数据集的实验室条件下准确率达到99.62%，在单任务油类识别中准确率达到100%，大幅超越了基线模型。这项工作证明了将LIF与基于先进优化的机器学习相结合对于在复杂结冰条件下进行可靠溢油检测的有效性。所提出的方法为寒冷和极地海洋区域的环境监测提供了一种可行的基于荧光的策略。
