Near-infrared Spectrum Detection of Fish Oil DHA Content Based on Empirical Mode Decomposition and Independent Component Analysis

Abstract

The near infrared (NIR) spectrum of fish oil is often very weak, and part of the peaks are submerged in the noise and are difficult to distinguish especially when NIR spectrum is applied to component analysis. A new method is proposed to get the pretreatment of NIR spectrum, which combines empirical mode decomposition (EMD) with independent component analysis (ICA). The principle and steps of method are given and its de-noising effect is evaluated by some parameters. With experiment, it is indicated that the de-noising effect of fish oil spectrum is slightly better than that of Wavelet and EMD method. After de-noising, the noise has been almost completely inhibited and the characteristic peak of spectrum is preserved well. SNR reaches 34.613. RMSE is only 0.00257 and SR reaches 0.99976. The horizontal feature and vertical features of spectrum are retained well. Then the fish oil DHA content is calculated based on the de-noised spectrum. The correlation ratio of the prediction set is improved to 0.9887 from 0.9682, and the RMSEP is reduced to 0.0308 from 0.0572. These improved that the proposed method is effective to get the pretreatment of NIR spectrum and improves the accuracy of near-infrared spectrum detection of fish oil DHA content.