Identification of surface-enhanced Raman spectroscopy using hybrid transformer network.

Surface-enhanced Raman Spectroscopy (SERS) is extensively implemented in drug detection due to its sensitivity and non-destructive nature. Deep learning methods, which are represented by convolutional neural network (CNN), have been widely applied in identifying the spectra from SERS for powerful learning ability. However, the local receptive field of CNN limits the feature extraction of sequential spectra for suppressing the analysis results. In this study, a hybrid Transformer network, TMNet, was developed to identify SERS spectra by integrating the Transformer encoder and the multi-layer perceptron. The Transformer encoder can obtain precise feature representations of sequential spectra with the aid of self-attention, and the multi-layer perceptron efficiently transforms the representations to the final identification results. TMNet performed excellently, with identification accuracies of 99.07% for the spectra of hair containing drugs and 97.12% for those of urine containing drugs. For the spectra with additive white Gaussian, baseline background, and mixed noises, TMNet still exhibited the best performance among all the methods. Overall, the proposed method can accurately identify SERS spectra with outstanding noise resistance and excellent generalization and holds great potential for the analysis of other spectroscopy data.

Rapid and accurate identification of bakanae pathogens carried by rice seeds based on hyperspectral imaging and deep transfer learning.

Bakanae disease is a common seed-borne disease of rice. Rapid and accurate detection of bakanae pathogens carried by rice seeds is essential for the health of rice germplasm resources and the safety of rice production. This study aims to propose a general framework for species identification of major bakanae pathogens carried by rice seeds based on hyperspectral imaging and deep transfer learning. Seven varieties of rice seeds and four kinds of bakanae pathogens were analyzed. One-dimensional deep convolution neural networks (DCNNs) were first constructed using complete datasets. They achieved accuracies larger than 96.5% on the testing sets of most datasets, exceeding the conventional SVM and PLS-DA models. Then the developed DCNNs were transferred to detect other complete training sets. Most of the deep transferred models achieved comparable or even better performance than the original DCNNs. Two smaller target training sets were further constructed by randomly selecting spectra from the complete training sets. As the size of the target training sets reduced, the accuracies of all models on the corresponding testing sets also decreased gradually. Visualization analysis were conducted using the t-distribution stochastic neighbor embedding (t-SNE) algorithm and a proposed gradient-weighted activation wavelength (Grad-AW) method. They all showed that deep transfer learning could utilize the representation patterns in the source datasets to improve the target tasks. The overall results indicated that the bakanae pathogens were all identified accurately under our proposed framework. Hyperspectral imaging combined with deep transfer learning provided a new idea for the quality detection of large-scale seeds in modern seed industry.

Acceleration of high-quality Raman imaging via a locality enhanced transformer network.

Raman imaging (RI) is an outstanding technique that enables molecular-level medical diagnostics and therapy assessment by providing characteristic fingerprint and morphological information about molecules. However, obtaining high-quality Raman images generally requires a long acquisition time, up to hours, which is prohibitive for RI applications of timely cytopathology and histopathology analyses. To address this issue, image super-resolution (SR) based on deep learning, including convolutional neural networks and transformers, has been widely recognized as an effective solution to reduce the time required for achieving high-quality RI. In this study, a locality enhanced transformer network (LETNet) is proposed to perform Raman image SR. Specifically, the general architecture of the transformer is adopted with the replacement of self-attention by convolution to generate high-fidelity and detailed SR images. Additionally, the convolution in the LETNet is further optimized by utilizing depth-wise convolution to improve the computational efficiency of the model. Experiments on hyperspectral Raman images of breast cancer cells and Raman images of a few channels of brain tumor tissues demonstrate that the LETNet achieves superior 2×, 4×, and 8× SR with fewer parameters compared with other SR methods. Consequently, high-quality Raman images can be obtained with a significant reduction in time, ranging from 4 to 64 times. Overall, the proposed method provides a novel, efficient, and reliable solution to expedite high-quality RI and promote its application in real-time diagnosis and therapy.

SERS with Flexible ß-CD@AuNP/PTFE Substrates for In Situ Detection and Identification of PAH Residues on Fruit and Vegetable Surfaces Combined with Lightweight Network.

The detection of polycyclic aromatic hydrocarbons (PAHs) on fruit and vegetable surfaces is important for protecting human health and ensuring food safety. In this study, a method for the in situ detection and identification of PAH residues on fruit and vegetable surfaces was developed using surface-enhanced Raman spectroscopy (SERS) based on a flexible substrate and lightweight deep learning network. The flexible SERS substrate was fabricated by assembling ß-cyclodextrin-modified gold nanoparticles (ß-CD@AuNPs) on polytetrafluoroethylene (PTFE) film coated with perfluorinated liquid (ß-CD@AuNP/PTFE). The concentrations of benzo(a)pyrene (BaP), naphthalene (Nap), and pyrene (Pyr) residues on fruit and vegetable surfaces could be detected at 0.25, 0.5, and 0.25 µg/cm2, respectively, and all the relative standard deviations (RSD) were less than 10%, indicating that the ß-CD@AuNP/PTFE exhibited high sensitivity and stability. The lightweight network was then used to construct a classification model for identifying various PAH residues. ShuffleNet obtained the best results with accuracies of 100%, 96.61%, and 97.63% for the training, validation, and prediction datasets, respectively. The proposed method realised the in situ detection and identification of various PAH residues on fruit and vegetables with simplicity, celerity, and sensitivity, demonstrating great potential for the rapid, nondestructive analysis of surface contaminant residues in the food-safety field.

Surface-enhanced Raman spectroscopy charged probes under inverted superhydrophobic platform for detection of agricultural chemicals residues in rice combined with lightweight deep learning network.

In this study, surface-enhanced Raman spectroscopy (SERS) charged probes and an inverted superhydrophobic platform were used to develop a detection method for agricultural chemicals residues (ACRs) in rice combined with lightweight deep learning network. First, positively and negatively charged probes were prepared to adsorb ACRs molecules to SERS substrate. An inverted superhydrophobic platform was prepared to alleviate the coffee ring effect and induce tight self-assembly of nanoparticles for high sensitivity. Chlormequat chloride of 15.5-0.05 mg/L and acephate of 100.2-0.2 mg/L in rice were measured with the relative standard deviation of 4.15% and 6.25%. SqueezeNet were used to develop regression models for the analysis of chlormequat chloride and acephate. And the excellent performances were obtained with the coefficients of determination of prediction of 0.9836 and 0.9826 and root-mean-square errors of prediction of 0.49 and 4.08. Therefore, the proposed method can realize sensitive and accurate detection of ACRs in rice.

Statistic and Network Features of RGB and Hyperspectral Imaging for Determination of Black Root Mold Infection in Apples.

Apples damaged by black root mold (BRM) lose moisture, vitamins, and minerals as well as carry dangerous toxins. Determination of the infection degree can allow for customized use of apples, reduce financial losses, and ensure food safety. In this study, red-green-blue (RGB) imaging and hyperspectral imaging (HSI) are combined to detect the infection degree of BRM in apple fruits. First, RGB and HSI images of healthy, mildly, moderately, and severely infected fruits are measured, and those with effective wavelengths (EWs) are screened from HSI by random frog. Second, the statistic and network features of images are extracted by using color moment and convolutional neural network. Meanwhile, random forest (RF), K-nearest neighbor, and support vector machine are used to construct classification models with the above two features of RGB and HSI images of EWs. Optimal results with the 100% accuracy of training set and 96% accuracy of prediction set are obtained by RF with the statistic and network features of the two images, outperforming the other cases. The proposed method furnishes an accurate and effective solution for determining the BRM infection degree in apples.

Detection of apple fruit damages through Raman spectroscopy with cascade forest.

Apple fruit damages seriously cause product and economic losses, infringe consumer rights and interests, and have harmful effects on human and livestock health. In this study, Raman spectroscopy (RS) and cascade forest (CForest) were adopted to determine apple fruit damages. First, the RS spectra of healthy, bruised, Rhizopus-infected, and Botrytis-infected apples were measured. Spectral changes and band attribution were analyzed. Different modeling methods were combined with various pre-processing and dimension reduction methods to construct recognition models. Among all models, CForest constructed with full spectra processed by Savitsky-Golay smoothing obtained the best performance with accuracies of 100%, 91.96%, and 92.80% in the training, validation, and test sets (ACCTE). And the modeling time is reduced to 1/3 of the full-spectra model with a similar ACCTE of 91.56% after principal component analysis. Overall, RS and CForest provided a non-destructive, rapid, and accurate identification of apple fruit damages and could be used in disease recognition and safety assurance of other fruits.

Drought stress identification of tomato plant using multi-features of hyperspectral imaging and subsample fusion.

Drought stress (DS) is one of the most frequently occurring stresses in tomato plants. Detecting tomato plant DS is vital for optimizing irrigation and improving fruit quality. In this study, a DS identification method using the multi-features of hyperspectral imaging (HSI) and subsample fusion was proposed. First, the HSI images were measured under imaging condition with supplemental blue lights, and the reflectance spectra were extracted from the HSI images of young and mature leaves at different DS levels (well-watered, reduced-watered, and deficient-watered treatment). The effective wavelengths (EWs) were screened by the genetic algorithm. Second, the reference image was determined by ReliefF, and the first four reflectance images of EWs that are weakly correlated with the reference image and mutually irrelevant were obtained using Pearson's correlation analysis. The reflectance image set (RIS) was determined by evaluating the superposition effect of reflectance images on identification. The spectra of EWs and the image features extracted from the RIS by LeNet-5 were adopted to construct DS identification models based on support vector machine (SVM), random forest, and dense convolutional network. Third, the subsample fusion integrating the spectra and image features of young and mature leaves was used to improve the identification further. The results showed that supplemental blue lights can effectively remove the high-frequency noise and obtain high-quality HSI images. The positive effect of the combination of spectra of EWs and image features for DS identification proved that RIS contains feature information pointing to DS. Global optimal classification performance was achieved by SVM and subsample fusion, with a classification accuracy of 95.90% and 95.78% for calibration and prediction sets, respectively. Overall, the proposed method can provide an accurate and reliable analysis for tomato plant DS and is hoped to be applied to other crop stresses.

Detection of amylase activity and moisture content in rice by reflectance spectroscopy combined with spectral data transformation.

In this study, reflectance spectroscopy was used to achieve rapid and non-destructive detection of amylase activity and moisture content in rice. Since rice husk can interfere with spectral measurements, spectral data transformation was used to remove the husk interference. Reflectance spectra of rice were transformed by direct standardization, convolutional autoencoder network, and kernel regression (KR). Then, random frog and elliptical envelope were adopted to select effective wavelengths, and partial least squares regression (PLSR) and support vector regression were used to establish analysis models. The optimal transformation was from KR, and PLSR and effective wavelengths of the transformed spectra obtained excellent performance with coefficient of determination of test of 0.6987 and 0.8317 and root-mean-square error of test of 0.3359 and 2.2239, respectively. The result was better than that of the rice spectra and was close to that of the husked rice spectra. When the moisture content was integrated into the regression model of amylase activity, a better result was obtained. Thus, the proposed method can detect amylase activity and moisture content in rice accurately.

Detection of 1-OHPyr in human urine using SERS with injection under wet liquid-liquid self-assembled films of ß-CD-coated gold nanoparticles and deep learning.

1-Hydroxypyrene (1-OHPyr), a typical hydroxylated polycyclic aromatic hydrocarbon (OH-PAH), has been commonly regarded as a urinary biomarker for assessing human exposure and health risks of PAHs. Herein, a fast and sensitive method was developed for the determination of 1-OHPyr in urine using surface-enhanced Raman spectroscopy (SERS) combined with deep learning (DL). After emulsification, urinary 1-OHPyr was separated using simple liquid-liquid extraction. Gold nanoparticles with ß-cyclodextrin (ß-CD@AuNPs) were synthesized, and homogeneous and ordered ß-CD@AuNP films were prepared through a liquid-liquid interface self-assembly process. The separated 1-OHPyr was injected under wet assembled films for SERS detection. Concentration as low as 0.05 µg mL-1 of 1-OHPyr in urine could still be detected, and the relative standard deviation was 5.5 %, and this was ascribed to the adsorption of ß-CD and the high-probability contact between 1-OHPyr molecules and the nanogap of assembled films under the action of capillary force. Meanwhile, a convolutional neural network (CNN), a classical DL network architecture, was adopted to build the prediction model, and the model was further simplified by genetic algorithm (GA). CNN combined with a GA obtained optimized results with determination coefficient and a root mean square error of prediction sets of 0.9639 and 0.6327, respectively, outperforming other models. Overall, the proposed method achieves fast and accurate detection of 1-OHPyr in urine, improves the assessment human exposure to PAHs and is expected to have applications in the analysis of other OH-PAHs in complex environments.

Detection of wheat Fusarium head blight using UAV-based spectral and image feature fusion.

Infection caused by Fusarium head blight (FHB) has severely damaged the quality and yield of wheat in China and threatened the health of humans and livestock. Inaccurate disease detection increases the use cost of pesticide and pollutes farmland, highlighting the need for FHB detection in wheat fields. The combination of spectral and spatial information provided by image analysis facilitates the detection of infection-related damage in crops. In this study, an effective detection method for wheat FHB based on unmanned aerial vehicle (UAV) hyperspectral images was explored by fusing spectral features and image features. Spectral features mainly refer to band features, and image features mainly include texture and color features. Our aim was to explain all aspects of wheat infection through multi-class feature fusion and to find the best FHB detection method for field wheat combining current advanced algorithms. We first evaluated the quality of the two acquired UAV images and eliminated the excessively noisy bands in the images. Then, the spectral features, texture features, and color features in the images were extracted. The random forest (RF) algorithm was used to optimize features, and the importance value of the features determined whether the features were retained. Feature combinations included spectral features, spectral and texture features fusion, and the fusion of spectral, texture, and color features to combine support vector machine, RF, and back propagation neural network in constructing wheat FHB detection models. The results showed that the model based on the fusion of spectral, texture, and color features using the RF algorithm achieved the best performance, with a prediction accuracy of 85%. The method proposed in this study may provide an effective way of FHB detection in field wheat.

Kernel principal component analysis and differential non-linear feature extraction of pesticide residues on fruit surface based on surface-enhanced Raman spectroscopy.

Surface-enhanced Raman spectroscopy (SERS) has attracted much attention because of its high sensitivity, high speed, and simple sample processing, and has great potential for application in the field of pesticide residue detection. However, SERS is susceptible to the influence of a complex detection environment in the detection of pesticide residues on the surface of fruits, facing problems such as interference from the spectral peaks of detected impurities, unclear dimension of effective correlation data, and poor linearity of sensing signals. In this work, the enhanced raw data of the pesticide thiram residues on the fruit surface using gold nanoparticle (Au-NPs) solution are formed into the raw data set of Raman signal in the IoT environment of Raman spectroscopy principal component detection. Considering the non-linear characteristics of sensing data, this work adopts kernel principal component analysis (KPCA) including radial basis function (RBF) to extract the main features for the spectra in the ranges of 653∼683 cm-1, 705∼728 cm-1, and 847∼872 cm-1, and discusses the effects of different kernel function widths (σ) to construct a qualitative analysis of pesticide residues based on SERS spectral data model, so that the SERS spectral data produce more useful dimensionality reduction with minimal loss, higher mean squared error for cross-validation in non-linear scenarios, and effectively weaken the interference features of detecting impurity spectral peaks, unclear dimensionality of effective correlation data, and poor linearity of sensing signals, reflecting better extraction effects than conventional principal component analysis (PCA) models.

Multi-scale convolution neural network with residual modules for determination of drugs in human hair using surface-enhanced Raman spectroscopy with a gold nanorod film self-assembled by inverted evaporation.

Detection of illegal drug users is crucial in controlling drug-related crimes, reducing drug prevalence, and protecting human lives to ensure social stability. In this study, surface-enhanced Raman spectroscopy (SERS) and deep learning networks were employed to determine methamphetamine, ketamine, and morphine in human hair. Drugs were obtained from hair through alkaline hydrolysis and liquid-liquid extraction, and gold nanorods were employed to prepare the self-assembled film as the SERS substrate by inverted evaporation. The film showed good uniformity and excellent sensitivity, with a relative standard deviation of 15.6% and a detection limit of at least 10-10 M in the SERS detection of crystal violet. The spectra of methamphetamine, ketamine, and morphine at 0.05-1.0, 0.1-2.0, and 0.1-2.0 ng/mg were obtained, and the three drugs could be detected. Inception, a multi-scale feature extraction network, was combined with residual modules (Inception-ResNet) to develop the identification models of drugs, and the effect of spectral input form as a vector or matrix was explored. Inception-ResNet with input form of matrix outweighed other methods with 100.00%, 100.00%, and 99.23% accuracies in the training, validation, and prediction sets, respectively. In brief, SERS and Inception-ResNet with the spectra in matrix form provide an efficient and accurate determination of drugs in human hair, enabling the retrospective evaluation of drug use, and the method will be anticipated to detect excitant, poison, and toxic chemicals in human hair.

Fusion of electronic nose and hyperspectral imaging for mutton freshness detection using input-modified convolution neural network.

Electronic nose (E-nose) and hyperspectral image (HSI) were combined to evaluate mutton total volatile basic nitrogen (TVB-N), which is a comprehensive index of freshness. The response values of 10 E-nose sensors were collected, and seven responsive sensors were screened via histogram statistics. Reflectance spectra and image features were extracted from HSI images, and the effective variables were selected through random frog and Pearson correlation analyses. With multi-source features, an input-modified convolution neural network (IMCNN) was constructed to predict TVB-N. The seven E-nose sensors, spectra of effective wavelengths (EWs), and five important image features were combined with IMCNN to achieve the best result, with the root mean square error, correlation coefficient, and ratio of performance deviation of the prediction set of 3.039 mg/100 g, 0.920, and 3.59, respectively. Hence, the proposed method furnishes an approach to accurately analyze mutton freshness and provide a technical basis for investigation of other meat qualities.

Raman Spectroscopy and Improved Inception Network for Determination of FHB-Infected Wheat Kernels.

Detection of infected kernels is important for Fusarium head blight (FHB) prevention and product quality assurance in wheat. In this study, Raman spectroscopy (RS) and deep learning networks were used for the determination of FHB-infected wheat kernels. First, the RS spectra of healthy, mild, and severe infection kernels were measured and spectral changes and band attribution were analyzed. Then, the Inception network was improved by residual and channel attention modules to develop the recognition models of FHB infection. The Inception-attention network produced the best determination with accuracies in training set, validation set, and prediction set of 97.13%, 91.49%, and 93.62%, among all models. The average feature map of the channel clarified the important information in feature extraction, itself required to clarify the decision-making strategy. Overall, RS and the Inception-attention network provide a noninvasive, rapid, and accurate determination of FHB-infected wheat kernels and are expected to be applied to other pathogens or diseases in various crops.

Determination of adulteration in wheat flour using multi-grained cascade forest-related models coupled with the fusion information of hyperspectral imaging.

Wheat flour (WF) is a common ingredient in staple foods. However, the presence of intentional or unintentional adulterants makes it difficult to guarantee WF quality. Multi-grained cascade forest (gcForest) model, a non-neural network deep learning structure, fused with image-spectra features from hyperspectral imaging (HSI) was employed for detecting adulterant type (peanut, walnut, or benzoyl peroxide) and the corresponding concentration (0.03%, 0.05%, 0.1%, 0.5%, 1%, and 2%). Based on the spectra of full wavelength and effective wavelength (EW) from hyperspectral images of WF samples, the gcForest-related models exhibited high performance (lowest ACCP = 92.45%) and stability (lowest area under the curve = 0.9986). Furthermore, the fusion of the EW and the image features extracted by the symmetric all convolutional neural network (SACNN) was used to establish the gcForest-related models. The maximum accuracy improvement of the fusion feature model relative to the single spectral model and the image model was 2.45% and 44.37%, respectively. The results indicate that the gcForest-related model, combined with the image-spectra fusion feature of HSI, provides an effective tool for detection in food and agriculture.

Reflectance spectroscopy with operator difference for determination of behenic acid in edible vegetable oils by using convolutional neural network and polynomial correction.

A novel polynomial correction method, order-adaptive polynomial correction (OAPC), was proposed to correct reflectance spectra with operator differences, and convolutional neural network (CNN) was used to develop analysis model to predict behenic acid in edible oils. With application of OAPC, CNN performed well with coefficient of determination of correction (R2cor) of 0.8843 and root mean square error of correction (RMSEcor) of 0.1182, outperforming partial least squares regression, support vector regression and random forest with OAPC, as well as the cases without OAPC. Based on 16 effective wavelengths selected by combination of bootstrapping soft shrinkage, random frog and Pearson's correlation, CNN and OAPC exhibited excellent performance with R2cor of 0.9560 and RMSEcor of 0.0730. Meanwhile, only 5% correction samples were selected by Kennard-Stone for OAPC. Overall, the proposed method could alleviate the impact of operator differences on spectral analysis, thereby providing potential to correct differences from measurement instruments or environments.

Reflectance Spectroscopy with Multivariate Methods for Non-Destructive Discrimination of Edible Oil Adulteration.

The visible and near-infrared (Vis-NIR) reflectance spectroscopy was utilized for the rapid and nondestructive discrimination of edible oil adulteration. In total, 110 samples of sesame oil and rapeseed oil adulterated with soybean oil in different levels were produced to obtain the reflectance spectra of 350-2500 nm. A set of multivariant methods was applied to identify adulteration types and adulteration rates. In the qualitative analysis of adulteration type, the support vector machine (SVM) method yielded high overall accuracy with multiple spectra pretreatments. In the quantitative analysis of adulteration rate, the random forest (RF) combined with multivariate scattering correction (MSC) achieved the highest identification accuracy of adulteration rate with the full wavelengths of Vis-NIR spectra. The effective wavelengths of the Vis-NIR spectra were screened to improve the robustness of the multivariant methods. The analysis results suggested that the competitive adaptive reweighted sampling (CARS) was helpful for removing the redundant information from the spectral data and improving the prediction accuracy. The PLSR + MSC + CARS model achieved the best prediction performance in the two adulteration cases of sesame oil and rapeseed oil. The coefficient of determination (RPcv2) and the root mean square error (RMSEPcv) of the prediction set were 0.99656 and 0.01832 in sesame oil adulterated with soybean oil, and the RPcv2 and RMSEPcv were 0.99675 and 0.01685 in rapeseed oil adulterated with soybean oil, respectively. The Vis-NIR reflectance spectroscopy with the assistance of multivariant analysis can effectively discriminate the different adulteration rates of edible oils.

Rapid Detection of Fatty Acids in Edible Oils Using Vis-NIR Reflectance Spectroscopy with Multivariate Methods.

The composition and content of fatty acids are critical indicators to identify the quality of edible oils. This study was undertaken to establish a rapid determination method for quality detection of edible oils based on quantitative analysis of palmitic acid, stearic acid, arachidic acid, and behenic acid. Seven kinds of oils were measured to obtain Vis-NIR spectra. Multivariate methods combined with pretreatment methods were adopted to establish quantitative analysis models for the four fatty acids. The model of support vector machine (SVM) with standard normal variate (SNV) pretreatment showed the best predictive performance for the four fatty acids. For the palmitic acid, the determination coefficient of prediction (RP2) was 0.9504 and the root mean square error of prediction (RMSEP) was 0.8181. For the stearic acid, RP2 and RMSEP were 0.9636 and 0.2965. In the prediction of arachidic acid, RP2 and RMSEP were 0.9576 and 0.0577. In the prediction of behenic acid, the RP2 and RMSEP were 0.9521 and 0.1486. Furthermore, the effective wavelengths selected by successive projections algorithm (SPA) were useful for establishing simplified prediction models. The results demonstrate that Vis-NIR spectroscopy combined with multivariate methods can provide a rapid and accurate approach for fatty acids detection of edible oils.