基于卷积神经网络的成品油红外光谱分类分析OA

For the environmental problems caused by the leakage and pollution of petroleum products in the process of transportation and use,the infrared spectrum for four kinds of refined oil products was classified and analyzed based on convolutional neural network(CNN),so as to trace the source of oil products leakage.In this paper,387 groups of infrared transmission spectra of four kinds of refined oil products and their mixtures were measured.Three methods including Savitzky-golay polynomial smoothing(S-G),standard normal transformation(SNV)and multiple scattering correction(MSC)were used to preprocess the spectral data.The classification models of CNN before and after preprocessing were established.respectively.The results showed that the accuracy of the CNN classification models established by the preprocessed spectral data was higher than that of the original data,and the spectral data preprocessed by SNV showed the best model classification accuracy of 0.974 4,with a loss value of 0.257 9.The results showed that the detection method based on CNN combined with infrared transmission spectrum was feasible for the classification of refined oil varieties,and it provided theoretical support for the subsequent realization of efficient and rapid detection of petroleum pollutants.