多尺度连续小波变换岸基高光谱溢油厚度定量监测  

Quantitative monitoring of oil spill thickness using multiscale continuous wavelet-based shore-based hyperspectral analysis

作　　者：董杨 崔厚欣 邓家春 马俊杰 秦海啸[1] DONG Yang;CUI Hou-xin;DENG Jia-chun;MA Jun-jie;QIN Hai-xiao(Institute of Disaster Prevention,Sanhe 065201,China;Hebei Sailhero Environmental Protection High-tech Co.,Ltd,Shijiazhuang 050035,China)

机构地区：[1]防灾科技学院,河北三河市065201 [2]河北先河环保科技股份有限公司,河北石家庄050035

出　　处：《中国环境科学》2024年第11期6397-6407,共11页China Environmental Science

基　　金：国家重点研发计划(2023YFE0102400);河北省高等教育教学改革研究与实践项目(2022GJJG490)。

摘　　要：聚焦于对0#柴油油膜厚度的监测,分析不同厚度油膜的光谱曲线特征;为深入挖掘光谱数据与油膜厚度间的复杂关系,引入Morlet多尺度连续小波变换(CWT)技术,精准筛选出与油膜厚度高度敏感的特征波段,成功解决高光谱数据维度高、处理复杂的难题,从而显著提升厚度回归预测的准确性.利用Cat Boost回归模型高效的计算能力、强大的特征捕捉能力以及优异的泛化能力来高效监测这些敏感特征,进而构建精确的油膜厚度回归预测模型,提升了溢油事件的实时监测速度,确保了预测结果的精确性,为溢油应急响应的迅速启动与精准防控策略的制定提供了参考.实验及现场验证结果表明:多尺度连续小波变换技术在本研究中发挥了关键作用,它能够有效地从海量的高光谱数据中提取出与油膜厚度高度相关的敏感波段,从而提升了溢油厚度监测的准确率和效率.Cat Boost回归模型能够更好地捕捉油膜厚度变化特征,进一步增强模型泛化能力和鲁棒性,Cat Boost回归模型建立的柴油油膜厚度预测模型,表现出了极高的精确度,其验证集上R^(2)=0.90,RMSE=95.14mm,d=30.126%.Focusing on the monitoring of 0#diesel oil film thickness,the primary objective is to meticulously analyze the spectral curve characteristics of different thickness oil films.Secondly,to delve deeper into the intricate relationship between spectral data and oil film thickness,the Morlet multi-scale continuous wavelet transform(CWT)technique is introduced.This enables the precise identification of spectral bands that are highly sensitive to oil film thickness,effectively addressing the challenge posed by the high dimensionality and complexity of hyperspectral data.Consequently,this approach significantly enhances the accuracy of thickness regression predictions.At the same time,the CatBoost regression model,with its efficient computing performance,strong feature capture ability,and excellent generalization ability,efficiently integrates these sensitive features and constructs a precise regression prediction model of the oil film thickness,accelerating the real-time monitoring speed of oil spill events,thereby achieving the immediate capture of changes in the thickness of the oil film and ensuring the accuracy of the prediction results,providing a solid scientific basis and technical support for the rapid initiation of oil spill emergency responses and the formulation of precise prevention and control strategies.The results show that the multi-scale continuous wavelet transform technology plays a key role in this study.It can effectively extract the sensitive bands highly related to the thickness of the oil film from the massive hyperspectral data,thereby significantly improving the accuracy and efficiency of oil spill thickness monitoring.The CatBoost regression model can better capture the change category characteristic data of the oil film thickness,further enhancing the generalization ability and robustness of the model.The diesel oil film thickness prediction model established by the CatBoost regression model shows extremely high accuracy,with R2=0.90,RMSE=95.14μm,δ=30.126%on the validation set.

关 键 词：高光谱成像 光谱分析 连续小波变换 溢油厚度 定量监测 

分 类 号：X55[环境科学与工程—环境工程]