ABSTRACT
This study investigated the feasibility of using hyperspectral imaging (HSI) technique to detect the saponin content in Panax notoginseng (PN) powder. The reflectance hyperspectral images of PN powder samples were collected in the spectral range of 400.6-999.9 nm. Savitzky-golay (SG) smoothing combined with detrending correction was utilized to preprocess the original spectral data. Two model population analysis (MPA) based methods, namely bootstrapping soft shrinkage (BOSS) and iteratively retains informative variables (IRIV) were employed to extract feature wavelengths from the full spectra. A generalized normal distribution optimization based extreme learning machine (GNDO-ELM) model was proposed to establish calibration model between spectra and saponin content, and compared with existing methods (GA-ELM, PSO-ELM and SSA-ELM). The result showed that the IRIV-GNDO-ELM model gave the best performance, with coefficient of determination for prediction (R2P) of 0.953 and root mean square error for prediction (RMSEP) of 0.115%. Therefore, it is possible to determine the saponin content of PN powder by using HSI technique.
Subject(s)
Panax notoginseng , Saponins , Hyperspectral Imaging , Powders , Least-Squares Analysis , AlgorithmsABSTRACT
Lipid and protein oxidation are the main causes of meat deterioration during freezing. Traditional methods using hyperspectral imaging (HSI) need to train multiple independent models to predict multiple attributes, which is complex and time-consuming. In this study, a multi-task convolutional neural network (CNN) model was developed for visible near-infrared HSI data (400-1002 nm) of 240 pork samples treated with different freeze-thaw cycles (0-9 cycles) to evaluate the feasibility of simultaneously monitoring lipid oxidation (thiobarbituric acid reactive substance content) and protein oxidation (carbonyl content) in pork. The performance of the commonly used partial least squares regression (PLSR) model based on the spectra after pre-processing (Standard normal variate, Savitzky-Golay derivative, and Savitzky-Golay smoothing) and feature selection (Regression coefficients) and single-output CNN model was compared. The results showed that the multi-task CNN model achieved the optimal prediction accuracies for lipid oxidation (R2p = 0.9724, RMSEP = 0.0227, and RPD = 5.2579) and protein oxidation (R2p = 0.9602, RMSEP = 0.0702, and RPD = 4.6668). In final, the changes of lipid and protein oxidation of pork in different freeze-thaw cycles were successfully visualized. In conclusion, the combination of HSI and multi-task CNN method shows the potential of end-to-end prediction of pork oxidative damage. This study provides a new, convenient and automated technique for meat quality detection in the food industry.
Subject(s)
Pork Meat , Red Meat , Animals , Swine , Red Meat/analysis , Spectroscopy, Near-Infrared/methods , Hyperspectral Imaging , Freezing , Oxidative Stress , Neural Networks, Computer , Least-Squares Analysis , LipidsABSTRACT
This study evaluated the feasibility of nondestructive testing and visualization of compound heavy metals (cadmium and lead) in lettuce leaves using fluorescence hyperspectral imaging. In addition, a method involving wavelet transform and stepwise regression (WT-SR) was proposed to perform dimensionality reduction of fluorescence spectral data. Fluorescent hyperspectral image acquisition and mathematical analysis were carried out on lettuce leaf samples processed with different compound heavy metal concentrations. The entire lettuce leaf sample was selected as a region of interest (ROI). Savitzky-Golay (SG) algorithm, multivariate scatter correction (MSC), standard normalized variable (SNV), first derivative (1st Der) and second derivative (2nd Der) were used to preprocess the ROI fluorescence spectra. Further, the successive projections algorithm (SPA), the competitive adaptive reweighted sampling (CARS), the iteratively retaining informative variables (IRIV) and variable iterative space shrinkage approach (VISSA), and the wavelet transform combined with stepwise regression (WT-SR) were used to reduce the dimension of spectral data. Finally, the multiple linear regression (MLR) algorithm was used to build the compound heavy metal content detection models. The results showed that the MLR models based on the feature data obtained by 1st Der-WT-SR achieved reasonable performance with Rp2 of 0.7905, RMSEP of 0.0269 mg/kg and RPD of 2.477 for Cd content under wavelet fifth layer decomposition, and with Rp2 of 0.8965, RMSEP of 0.0096 mg/kg and RPD of 3.211 for Pb content under wavelet first layer decomposition. The distribution maps of cadmium and lead contents in lettuce leaves were established using the optimal prediction models. The results further confirmed the great potential of fluorescence hyperspectral technology combined with optimization algorithm for the detection of compound heavy metals.
Subject(s)
Cadmium , Metals, Heavy , Cadmium/analysis , Lactuca , Hyperspectral Imaging , Least-Squares Analysis , Algorithms , Plant Leaves/chemistryABSTRACT
The evaluation capability of hyperspectral imaging technology was studied for the forecasts of heavy metal lead concentration of oilseed rape plant. In addition, a transfer stacked auto-encoder (T-SAE) algorithm including two network methods, the dual-model T-SAE and the single-model T-SAE, was proposed in this paper. The hyperspectral images of oilseed rape leaf and root were acquired under different Pb stress concentrations. The entire region of the oilseed rape leaf (or root) was selected as the region of interest (ROI) to extract the spectral data, and standard normalized variable (SNV), first derivative (1st Der) and second derivative (2nd Der) were used to preprocess the ROI spectra. Besides, the principal component analysis (PCA) algorithm was used to reduce the dimensionality of the spectral data before and after preprocessing. Hence, the best pre-processed data was determined for subsequent research and analysis. Furthermore, the SAE deep learning networks were built based on the oilseed rape leaf data, oilseed rape root data, and the combined data of oilseed rape leaf and root based on the best pre-processed spectral data. Finally, the T-SAE models were obtained through transfer learning of the best SAE deep learning network. The results show that the best preprocessing algorithms of the oilseed rape leaf and root spectra were SNV and 1st Der algorithm, respectively. In addition, the prediction set recognition accuracy of the best T-SAE model of Pb stress gradient in oilseed rape plants was 98.75%. Additionally, the prediction set coefficient of determination of the best T-SAE model of the Pb content in the oilseed rape leaf and root data were 0.9215 and 0.9349, respectively. Therefore, a deep transfer learning method combined with hyperspectral imaging technology can effectively realize the the qualitative and quantitative detection of heavy metal Pb in oilseed rape plants.
Subject(s)
Brassica napus , Metals, Heavy , Lead , Hyperspectral Imaging , Least-Squares Analysis , Plant Leaves , Vegetables , Technology , Machine LearningABSTRACT
The purpose of this study was to develop a deep learning method involving wavelet transform (WT) and stacked denoising autoencoder (SDAE) for extracting deep features of heavy metal lead (Pb) detection of oilseed rape leaves. Firstly, the standard normalized variable (SNV) algorithm was established as the best preprocessing algorithm, and the SNV-treated fluorescence spectral data was used for further data analysis. Then, WT was used to decompose the SNV-treated fluorescence spectra of oilseed rape leaves to obtain the optimal wavelet decomposition layers using different wavelet basis functions, and SDAE was used for deep feature learning under the optimal wavelet decomposition layer. Finally, the best established support vector machine regression (SVR) model prediction set parameters Rp2, RMSEP and RPD were 0.9388, 0.0199 mg/kg and 3.275 using sym7 as the wavelet basis function. The results of this study verified that the huge potential of fluorescence hyperspectral technology combined with deep learning algorithms to detect heavy metals.
Subject(s)
Brassica napus , Deep Learning , Spectroscopy, Near-Infrared/methods , Hyperspectral Imaging , Least-Squares Analysis , Plant Leaves , AlgorithmsABSTRACT
BACKGROUND: Oilseed rape, as one of the most important oil crops, is an important source of vegetable oil and protein for mankind. As a non-essential element for plant growth, heavy metal cadmium (Cd) is easily absorbed by plants. Cd will inhibit the photosynthesis of plants, destroy the cell structure, slow the growth of plants, and affect their development and yield. It is necessary to develop a method based on visible near-infrared (NIR) hyperspectral imaging (HSI) technology to quickly and nondestructively determine the Cd content in rape leaves. RESULTS: Two-layer estimation models were established by combining visible-NIR HSI with ensemble learning methods (stacking and blending). One layer used support vector regression, extreme learning machine, decision tree, and random forest (RF) as basic learners, and the other layer used support vector regression or RF as a meta learner. Different models were used to analyze the spectra of rape treated with five Cd concentrations to obtain the best prediction method. The results showed that the best model to predict Cd content was the stacking ensemble model with RF as the meta learner, with coefficient of determination for prediction of 0.9815 and root-mean-square error for prediction of 5.8969 mg kg-1 . A pseudo-color image was developed using this stacking model to visualize the content and distribution of Cd. CONCLUSION: The combination of visible-NIR HSI technology and the stacking ensemble learning method is a feasible method to detect the Cd content in rape leaves, which has the potential of being rapid and nondestructive. © 2022 Society of Chemical Industry.
Subject(s)
Brassica rapa , Cadmium , Cadmium/analysis , Least-Squares Analysis , Support Vector Machine , Plant Oils/chemistry , Plant Leaves/chemistry , VegetablesABSTRACT
This study evaluated the feasibility of non-destructive detection of carbonyl and total sulfhydryl contents by fluorescence hyperspectral imaging (F-HSI) to visualize the protein oxidation degree of pork during freezing-thawing process. Fluorescence hyperspectral image acquisition and chemical analysis were carried out on pork samples treated with different freeze-thaw cycles. Variational Mode Decomposition (VMD) was used to preprocess the raw spectrum, and Mutual Information-Variance Inflation Factor (MI-VIF) was applied to select the feature wavelengths. The Partial least squares regression (PLSR) models based on selected 19 wavelengths obtained good performance in predicting carbonyl content with R2p of 0.9275 and RMSEP of 0.0812 nmol/mg, and sulfhydryl content with R2p of 0.9512 and RMSEP of 1.2979 nmol/mg. The distribution maps of carbonyl and total sulfhydryl content were established based on the optimal prediction models. The results confirmed that the contents of carbonyl and total sulfhydryl in pork could be successfully predicted by F-HSI, so as to monitor the protein oxidation degree of pork during freezing-thawing.
Subject(s)
Pork Meat , Red Meat , Animals , Swine , Red Meat/analysis , Spectroscopy, Near-Infrared/methods , Pork Meat/analysis , Freezing , Hyperspectral Imaging , Least-Squares AnalysisABSTRACT
S-ovalbumin content is an indicator of egg freshness and has an important impact on the quality of processed foods. The objective of this study is to develop simplified models for monitoring the S-ovalbumin content of eggs during storage using hyperspectral imaging (HSI) and multivariate analysis. The hyperspectral images of egg samples at different storage periods were collected in the wavelength range of 401-1002 nm, and the reference S-ovalbumin content was determined by spectrophotometry. The standard normal variate (SNV) was employed to preprocess the raw spectral data. To simplify the calibration models, competitive adaptive reweighted sampling (CARS) was applied to select feature wavelengths from the whole spectral range. Based on the full and feature wavelengths, partial least squares regression (PLSR) and least squares support vector machine (LSSVM) models were developed, in which the simplified LSSVM model yielded the best performance with a coefficient of determination for prediction (R2P) of 0.918 and a root mean square error for prediction (RMSEP) of 7.215%. By transferring the quantitative model to the pixels of hyperspectral images, the visualizing distribution maps were generated, providing an intuitive and comprehensive evaluation for the S-ovalbumin content of eggs, which helps to understand the conversion of ovalbumin into S-ovalbumin during storage. The results provided the possibility of implementing a multispectral imaging technique for online monitoring the S-ovalbumin content of eggs.
ABSTRACT
Zinc (Zn) content plays a decisive role in plant growth. Accurate management of Zn fertilizer application can promote high-quality development of the oilseed rape industry. This study adopted a deep learning (DL) method to predict the Zn content of oilseed rape leaves using hyperspectral imaging (HSI). The dropout mechanism was introduced to improve the stacked sparse autoencoder (SSAE) and named modified SSAE (MSSAE). MSSAE extracted deep spectral features of samples based on pixel-level spectral information (the wavelength range of the spectrum is 431-962 nm). Subsequently, the deep spectral features were applied as the inputs for support vector regression (SVR) and least squares support vector regression (LSSVR) to predict the Zn content in oilseed rape leaves. In addition, the successive projections algorithm (SPA) and the variable iterative space shrinkage approach (VISSA) were investigated as wavelength selection algorithms for comparison. The results showed that the MSSAE-LSSVR model had the best prediction performance (the coefficient of determination (R2) and root mean square error (RMSE) of the prediction set were 0.9566 and 1.0240 mg/kg, respectively). The overall results showed that the MSSAE was able to extract the deep features of HSI data and validated the possibility of HSI combined with a DL method for nondestructive testing of Zn content in oilseed rape leaves.
Subject(s)
Brassica napus , Hyperspectral Imaging , Algorithms , Least-Squares Analysis , Plant Leaves , Support Vector Machine , Vegetables , ZincABSTRACT
Exploring the cadmium (Cd) pollution in rape is of great significance to food safety and consumer health. In this study, a rapid, nondestructive and accurate method for the determination of Cd content in rape leaves based on hyperspectral imaging (HSI) technology was proposed. The spectral data of rape leaves under different Cd stress from 431 nm to 962 nm were collected by visible-near infrared HSI spectrometer. In order to improve the robustness and accuracy of the regression model, a machine learning algorithm was proposed, named multi-disturbance bagging Extreme Learning Machine (MdbaggingELM). The prediction models of Cd content in rape leaves based on MdbaggingELM and ELM-based method (ELM and baggingELM) were established and compared. The results showed that the model of the proposed MdbaggingELM method performed significantly in the prediction of Cd content, with Rp2 of 0.9830 and RMSEP of 2.8963 mg/kg. The results confirmed that MdbaggingELM is an efficient regression algorithm, which played a positive role in enhancing the stability and the prediction effect of the model. The combination of MdbaggingELM and HSI technology has great potential in the detection of Cd content in rape leaves.
Subject(s)
Cadmium , Hyperspectral Imaging , Algorithms , Least-Squares Analysis , Plant Leaves , Technology , VegetablesABSTRACT
Feature selection plays a vital role in the quantitative analysis of high-dimensional data to reduce dimensionality. Recently, the variable selection method based on mutual information (MI) has attracted more and more attention in the field of feature selection, where the relevance between the candidate variable and the response is maximized and the redundancy of the selected variables is minimized. However, multicollinearity often is a serious problem in linear models. Collinearity can cause unstable parameter estimation, unreliable models, and weak predictive ability. In order to address this problem, the variance inflation factor (VIF) was introduced for feature selection. Therefore, a variable selection method based on MI combined with VIF was proposed in this paper, called Mutual Information-Variance Inflation Factor (MI-VIF). By calculating the MI between the independent variable and the response variable, the variable with greater MI was selected to maximize the correlation between the independent variable and the response variable. By calculating the VIF between the independent variables, the multicollinearity test was performed. The variables that cause the multicollinearity of the model were eliminated to minimize the collinearity between the independent variables. The proposed method was tested based on two high-dimensional spectral datasets. The regression models (PLSR, MLR) were established based on feature selection through MI-VIF and MI-based methods (MIFS, MMIFS) to compare the prediction accuracy of the models. The results showed that under two datasets, the MI-VIF showed a good prediction performance. Based on the tea dataset, the established MI-VIF-MLR model achieved accuracy with Rp2 of 0.8612 and RMSEP of 0.4096, the MI-VIF-PLSR model achieved accuracy with Rp2 of 0.8614 and RMSEP of 0.4092. Based on the diesel fuels dataset, the established MI-VIF-MLR model achieved accuracy with Rp2 of 0.9707 and RMSEP of 0.6568, the MI-VIF-PLSR model achieved accuracy with Rp2 of 0.9431 and RMSEP of 0.9675. In addition, the MI-VIF was compared with the Successive projections algorithm (SPA), which is a method to reduce the collinearity between variables in the wavelength selection of the near-infrared spectrum. It was found that MI-VIF also had a good predictive effect compared to SPA. It proves that the MI-VIF is an effective variable selection method.
Subject(s)
Algorithms , Spectroscopy, Near-Infrared , Gasoline , Least-Squares Analysis , Linear ModelsABSTRACT
The feasibility analysis of fluorescence hyperspectral imaging technology was studied for the detection of lead content in lettuce leaves. Further, Monte Carlo optimized wavelet transform stacked auto-encoders (WT-MC-SAE) was proposed for dimensionality reduction and depth feature extraction of fluorescence spectral data. The fluorescence hyperspectral images of 2800 lettuce leaf samples were selected and the whole lettuce leaf was used as the region of interest (ROI) to extract the fluorescence spectrum. Five different pre-processing algorithms were used to pre-process the original ROI spectral data including standard normalized variable (SNV), first derivative (1st Der), second derivative (2ndDer), third derivative (3rd Der) and fourth derivative (4th Der). Moreover, wavelet transform stacked auto-encoders (WT-SAE) and WT-MC-SAE were used for data dimensionality reduction, and support vector machine regression (SVR) was used for modeling analysis. Among them, 4th Der tends to be the most useful fluorescence spectral data for Pb content detection at 0.067 â¼ 1.400 mg/kg in lettuce leaves, with Rc2 of 0.9802, RMSEC of 0.02321 mg/kg, Rp2 of 0.9467, RMSEP of 0.04017 mg/kg and RPD of 3.273, and model scale (the number of nodes in the input layer, hidden layer and output layer) was 407-314-286-121-76 under the fifth level of wavelet decomposition. Further studies showed that WT-MC-SAE realizes the depth feature extraction of the fluorescence spectrum, and it is of great significance to use fluorescence hyperspectral imaging to realize the quantitative detection of lead in lettuce leaves.
Subject(s)
Deep Learning , Metals, Heavy , Algorithms , Lead , Least-Squares Analysis , Lactuca , Plant Leaves , TechnologyABSTRACT
Total soluble solids (TSS) are one of the most essential attributes determining the quality and price of fruit. This study aimed to use hyperspectral imaging (HSI) and wavelength selection for nondestructive detection of TSS in grape. A novel method involving variational mode decomposition and regression coefficients (VMD-RC) was proposed to select feature wavelengths. VMD was introduced to decompose the hyperspectral images data of samples with bands of (400.68-1001.61 nm) to get a series of feature components. Afterward, these components were processed separately using regression analysis to obtain the stability values of RC of each component. Finally, a filter-based selection strategy was used to screen key wavelengths. The least squares support vector machine (LSSVM) and partial least squares (PLS) were constructed under full and feature wavelengths for predicting TSS. The VMD-RC-LSSVM model obtained the best prediction accuracy for TSS, with R p 2 $R_p^2$ of 0.93, with R M S E P $RMSEP$ of 0.6 %, with R E R $RER$ of 18.14 and R P D p $RP{D_p}$ of 3.69. The overall results indicated that the VMD-RC algorithm can be used as an alternative to handle high-dimensional hyperspectral images data, and HSI has great potential for nondestructive and rapid evaluation of quality attributes in fruit. PRACTICAL APPLICATION: Traditional methods of evaluating grape quality attributes are destructive, time-consuming and laborious. Therefore, HSI was used to achieve rapid and nondestructive determination of TSS in grape. The results indicated that it was feasible to use HSI and variable selection for predicting TSS. It is of great significance to improve grape quality, guide postharvest handling and provide a valuable reference for noninvasively evaluating other internal attributes of fruit.