Moving Average Smoothing Research Articles

The qualitative and quantitative analysis of industrial paraffin contamination levels in rice using spectral pretreatment combined with machine learning models

Several bad traders have added industrial paraffin to rice in recent years to obtain more benefits. However, eating such adulterated rice poses a hazard to people's health, so the ability to detect the industrial paraffin contamination levels (IPCL) in rice is crucial. This study investigated the effects of spectral pretreatment combined with machine learning models on the qualitative and quantitative models of rice IPCL. Multiplicative scatter correction (MSC), standard normal variate (SNV), normalisation (NL), moving average smoothing (MAS), Savitzky-Golay smoothing (S-G-S), first derivative (FD) and second derivatives (SD) were used to preprocess the entire spectral range original spectral data. The data were then compared and analysed using PCA. The rice IPCL was detected using a qualitative model (SVM) and quantitative models (LSSVR, PSO-LSSVR, 1D-CNN). The results showed that NL-SVM (OA: 99.383%, Kappa: 0.993) and MAS-1D-CNN (R2: 0.964, RMSE: 0.033%) achieved the best detection results respectively. In addition, the NL-SVM (OA: 97.531%, Kappa: 0.970) and MAS-1D-CNN (R2: 0.926, RMSE: 0.040%) models after lightweight processing by CARS algorithm still showed good detection results. This study has a positive effect on the detection of rice IPCL, and contributed to the development of non-destructive testing technology for other agricultural and sideline products globally.

Nondestructive determination of IMP content in chilled chicken based on hyperspectral data combined with chemometrics

This study was conducted to investigate the potential of hyperspectral imaging technique (900-1700 nm) for nondestructive determination of inosinic acid (IMP) in chicken. Hyperspectral images of chicken flesh samples were acquired, and their mean spectra within the images were extracted. The quantitative relationship between the mean spectra and reference IMP value was fitted by partial least squares (PLS) regression algorithm. A PLS model (MAS-PLS) built with moving average smoothing (MAS) spectra showed better performance in predicting IMP content, leading to correlation coefficients (RP) of 0.951, root mean square error (RMSEP) of 0.046 mg/g, and residual predictive deviation (RPD) of 3.152. Regression coefficient (RC), successive projections algorithm (SPA), stepwise, competitive adaptive reweighted sampling (CARS), and uninformative variable elimination (UVE) were used to select the optimal wavelengths to optimize the MAS-PLS model. Based on the 18 optimal wavelengths (907.14, 917.02, 918.67, 926.90, 930.20, 936.78, 956.54, 1004.28, 1135.89, 1211.56, 1302.07, 1367.94, 1397.60, 1488.31, 1680.17, 1683.49, 1686.80 and 1695.10 nm) selected from MAS spectra by SPA, the MAS-SPA-PLS model was built with RP of 0.920, RMSEP of 0.056 mg/g and RPD of 3.220, which was similar to the MAS-PLS model. The overall study indicated that hyperspectral imaging in the 900-1700 nm range combined with PLS and SPA could be used to predict the IMP content in chicken flesh. Keywords: near-infrared hyperspectral imaging, chicken, inosinic acid, partial least squares, successive projections algorithm DOI: 10.25165/j.ijabe.20221501.6612 Citation: Wang Y Y, He H J, Jiang S Q, Ma H J. Nondestructive determination of IMP content in chilled chicken based on hyperspectral data combined with chemometrics. Int J Agric & Biol Eng, 2022; 15(1): 277–284.

Fast, simultaneous and contactless assessment of intact mango fruit by means of near infrared spectroscopy

This study aims to apply near infrared technology as a fast, simultaneous and non-destructive method for quality assessment on intact mango fruit in form of total soluble solids (TSS) and vitamin C. Absorbance spectra of 186 intact mango fruits with four different cultivars were acquired and recorded in wavelength ranging from 1000–2500 nm. Spectra data were enhanced and corrected using three different methods namely moving average smoothing (MAS), extended multiplicative scatter correction (EMSC) and standard normal variate (SNV). In addition, they were divided into two datasets namely calibration (n = 143) and prediction (n = 43) datasets consisting all four mango cultivars. The models used to predict TSS and vitamin C were developed using partial least square regression (PLSR). Prediction performance were quantified using correlation coefficient (r), root mean square error (RMSE), ratio prediction to deviation (RPD) and range to error ratio (RER) indexes. The results showed that the best prediction models for TSS and vitamin C were achieved when the models were constructed using EMSC correction approach with r = 0.86, RMSE = 1.67 Brix, RPD = 2.34 and RER = 9.72 for TSS. Meanwhile, for vitamin C, r = 0.86, RMSE = 6.84 mg·100g⁻¹, RPD = 2.00 and RER = 8.87. From this study, it was concluded that near infrared technology combined with proper spectra enhancement method may be applied as a rapid, simultaneous and contactless method for quality assessment on intact mangoes.

A Study on the Quantitative Viscosity Detection of Automobile Lubricating Oils Based on Near-infrared Spectroscopy

A quantitative viscosity prediction of automobile lubricating oils was performed. The near-infrared spectra were preprocessed by using three different techniques, namely, multiple scattering correction (MSC), moving average smoothing (MAS), and Savitzky–Golay smoothing (SG). The characteristic wavelength was extracted by using three different techniques, namely, successive projections algorithm (SPA), synergy interval partial least squares (siPLS), and interval partial least squares (iPLS). The initial data of the viscosity of lubricating oils collected by near-infrared (NIR) spectroscopy were preprocessed by using two different optimization methods, namely, particle swarm optimization (PSO) and genetic algorithm (PA). Finally, the quantitative viscosity detection model of the lubricating oils was built by combining with support vector regression (SVR). The optimal technical route for modeling was explored.

Noise reduction in the spectral domain of hyperspectral images using denoising autoencoder methods

Denoising of spectra has been a great challenge in hyperspectral image analysis. Near-infrared hyperspectral images of milk powder, rice flour and soybean flour were acquired and denoising in the spectral domain were studied. Noise free spectra and noises were simulated based on sample pixel-wise spectra. The noisy spectra with signal to noise ratio (SNR) around 45 ​dB (similar to real pixel-wise spectra) were simulated. The simulated noisy spectra were preprocessed by traditional methods as moving average smoothing (MAS), Savitzky-Golay smoothing (SGS), wavelet transform (WT) and empirical mode decomposition (EMD). The basic denoising autoencoder (DAE-1) and the stacked DAE (DAE-2) were studied for denoising. The noisy spectra with SNR around 35 ​dB and 55 ​dB were further simulated to explore the effectiveness of DAE based methods. DAE-1 and DAE-2 performed better than the other methods, with higher SNR, lower mean squared error (MSE) and mean absolute error (MAE). The developed DAE methods were applied to real-world pixel-wise spectra with good performances. The overall results proved the feasibility of using DAE based methods for noise reduction in the spectral domain of hyperspectral images, and the DAE based methods have great potential to be extended to spectral denoising of other vibrational spectroscopy techniques.

Handling of uncertainty due to interference fringe in FT-NIR transmittance spectroscopy — Performance comparison of interference elimination techniques using glucose-water system

The applicability of two elimination techniques for interferences occurring in measurements with cells of short pathlength using Fourier transform near-infrared (FT-NIR) spectroscopy was evaluated. Due to the growing interest in the field of vibrational spectroscopy in aqueous biological fluids (e.g. glucose in blood), aqueous solutions of d-(+)-glucose were prepared and split into a calibration set and an independent validation set. All samples were measured with two FT-NIR spectrometers at various spectral resolutions. Moving average smoothing (MAS) and fast Fourier transform filter (FFT filter) were applied to the interference affected FT-NIR spectra in order to eliminate the interference pattern. After data pre-treatment, partial least squares regression (PLSR) models using different NIR regions were constructed using untreated (interference affected) spectra and spectra treated with MAS and FFT filter. The prediction of the independent validation set revealed information about the performance of the utilized interference elimination techniques, as well as the different NIR regions. The results showed that the combination band of water at approx. 5200 cm−1 is of great importance since its performance was superior to the one of the so-called first overtone of water at approx. 6800 cm−1. Furthermore, this work demonstrated that MAS and FFT filter are fast and easy-to-use techniques for the elimination of interference fringes in FT-NIR transmittance spectroscopy.

Determination of Nitrogen Concentration in Fresh Pear Leaves by Visible/Near‐Infrared Reflectance Spectroscopy

A rapid and reliable method is required to determine the N status of pear (Pyrus communis L.) leaves during the growing season for timely fertilization to improve the yields and fruit quality. In the present study, we evaluated visible and near‐infrared reflectance (Vis/NIR) spectra of fresh pear leaves using partial least squares (PLS) regression to determine the N concentration of fresh pear leaves. In addition, we studied the performance of modified spectra generated using different preprocessing techniques. A total of 450 leaf samples were collected from 6‐yr‐old pear trees of two cultivars, and randomly separated into two subsets (calibration subset [294 samples] and validation subset [180 samples]) after excluding outliers by using principle component analysis. Results showed that the model built using full spectra performed better than that developed using characteristic wavelength segments. In addition, we found that original spectral proved to provide better accuracy than derivative spectra. Among the studied preprocessing techniques, moving average smoothing (MAS) technique improved accuracy the most. Overall results suggested that PLS regression with preprocessing of full spectra using MAS is optimal method for modeling N concentration of fresh pear leaves which yielded 0.961 and 0.953 coefficient of determination (R2) for calibration and cross‐validation, respectively. The validation of this method resulted high R2 value (0.847) and low mean relative error (4.48%). In conclusion, this model could provide a rapid and more reliable method to determine the total N concentration in fresh pear leaves and could be useful for fertilization management in pear orchards.

Trend analysis and short-term forecast of incident HIV infection in Ghana

The study uses time-series modelling to determine and predict trends in incident HIV infection in Ghana among specific age groups. The HIV data for Ghana were grouped according to northern and southern spatial sectors as they exhibited slightly different data collection formats. The trend of the epidemic is modelled using moving-average smoothing techniques, and the Box-Jenkins ARIMA model is used to forecast cases of newly acquired (incident) HIV infection. Trend analysis of past growth patterns reveals an increase in new cases of HIV infection in the northern sector, with the greatest increase occurring among persons aged 30 years and over. The epidemic in the southern sector appears to have levelled off. However, incident HIV infection in the 20–39-year-old age group of females in the sector is estimated to increase in the next three years. Moreover, the estimates suggest a higher increase in incident cases than that predicted by the National AIDS Control Programme. Nevertheless, incident HIV infection among persons aged 19 and below is found to be relatively stable. Thus, if efforts are made to reduce or prevent an increase in the number of new infections in the northern sector, and for the 20–39 years age group in the southern sector, Ghana will have a brighter future with regard to its response to the HIV epidemic. These findings can assist with developing strategic-intervention policy planning for Ghana and other countries in sub-Saharan Africa.