基于离散小波变换算法和Inception卷积模块一维卷积神经网络的石油类污染物红外光谱定性分析  

Classification Method for Petroleum Pollutants Based on Inception-One-Dimensional Convolutional Neural Network and Infrared Spectroscopy

作　　者：孔德明[1] 何绍炜 李心怡 赵珺瑜 宁晓东 KONG De-Ming;HE Shao-Wei;LI Xin-Yi;ZHAO Jun-Yu;NING Xiao-Dong(School of Electrical Engineering,Yanshan University,Qinhuangdao066000,China;School of Information Science and Engineering,YanshanUniversity,Qinhuangdao 066000,China)

机构地区：[1]燕山大学电气工程学院,秦皇岛066000 [2]燕山大学信息科学与工程学院,秦皇岛066000

出　　处：《分析化学》2024年第9期1287-1297,共11页Chinese Journal of Analytical Chemistry

基　　金：国家自然科学基金项目(No.62173289)资助。

摘　　要：红外光谱技术具有高效和无损等优点,在石油类污染物分类检测领域中具有重要的研究与应用价值。本研究提出了一种结合离散小波变换(DWT)算法和基于Inception卷积模块的一维卷积神经网络(Inception-1D-CNN)的石油类污染物分类方法,首先使用DWT算法对原始红外光谱数据进行去噪处理,消除因实验环境、仪器误差和人工操作等因素产生的干扰信息;再通过Inception-1D-CNN模型获取多尺度的红外光谱特征信息,并基于此模型对石油类污染物进行分类预测。实验结果表明,与标准正态变换(SNV)、迭代自适应加权惩罚最小二乘法(AirPLS)和卷积平滑(S-G)预处理方法相比,DWT算法结合卷积核大小为3×1的1D-CNN模型的预测准确率为86.6%,分别提高了6.6%、6.6%和3.3%;DWT算法结合卷积核大小为5×1的1D-CNN模型的预测准确率为93.3%,分别提高了10.0%、7.0%和3.3%;DWT算法结合卷积核大小为7×1的1D-CNN模型的预测准确率为90.0%,分别提高了6.7%、10.0%和3.4%;DWT算法结合Inception-1D-CNN模型的预测准确率为100.0%,分别提高了10.0%、10.0%和3.4%。因此,结合DWT算法和Inception-1D-CNN模型能够对石油类污染物准确分类预测,为后续海面溢油污染治理提供了一定的基础。Infrared spectroscopy technology has many advantages such as high efficiency and nondestructiveness,and has an important research and application value in the field of petroleum pollutant classification and detection.In this study,a petroleum pollutant classification method by combing the discrete wavelet transform(DWT)algorithm and a one-dimensional convolutional neural network based on the Inception module(Inception-1D-CNN)was proposed.Firstly,the DWT algorithm was used to denoise the original infrared spectral data to eliminate the interference information caused by experimental environment,instrument error and manual operation.Then,the inception-1D-CNN model was used to obtain multi-scale infrared spectroscopy feature information,and then classify the petroleum pollutants.Experimental results showed that compared with preprocessing methods such as standard normal variable(SNV),adaptive iteratively reweighted penalized least squares(AirPLS),and Savitzky-Golay smoothing(S-G),the prediction accuracy of the DWT algorithm combined with the 1D-CNN model with a convolutional kernel size of 3×1 was 86.6%,which was 6.6%,6.6%and 3.3%higher,respectively.The prediction accuracy of DWT algorithm combined with 1D-CNN model with a convolutional kernel size of 5×1 was 93.3%,which was 10.0%,7.0%and 3.3%higher,respectively.The prediction accuracy of the DWT algorithm combined with the 1D-CNN model with a convolutional kernel size of 7×1 was 90.0%,which was 6.7%,10.0%and 3.4%higher,respectively.The prediction accuracy of the DWT algorithm combined with the inception-1D-CNN model was 100.0%,which was 10.0%,10.0%and 3.4%higher,respectively.Therefore,the DWT algorithm combined with the inception-1D-CNN model could accurately classify and predict petroleum pollutants,and provided a certain basis for the subsequent treatment of oil spills on the sea surface.

关 键 词：红外光谱 石油类污染物 Inception模块 卷积神经网络 离散小波变换 

分 类 号：X834[环境科学与工程—环境工程] O657.33[理学—分析化学] TP183[理学—化学]