Classification of cow milk using artificial neural network developed from the spectral data of single- and three-detector spectrophotometers

Fluorescence spectroscopy combined with multilayer perceptron deep learning to identify the authenticity of monofloral honey-Rape honey

Honey authenticity is critical to honey quality. The development of a quick, easy, and non-destructive technique for determining the authenticity of honey encourages an improvement in honey quality. Here, the authenticity of monofloral honey-rape honey was determined using fluorescence spectroscopy combined with multilayer perceptron (MLP) deep learning, without the need for any prior feature extraction or pre-processing. A total of 91 raw fluorescence intensity data of the real and adulterated honey samples at a fixed excitation wavelength of 280 nm were first matrixed, and all data were then categorized into a training set, a validation set, and a test set with numbers of 64, 16, and 11, respectively. The connection with dropout was selected to build and link the MLP internal network. The activation function, learning rate, optimizer, and number of epochs were among the hyperparameters of the MLP neural network that were tuned. A good MLP deep learning network model for determining the authenticity of monofloral honey, rape honey, was developed after constant validation and debugging. According to the accuracy curve of the MLP model, the accuracy of the training set increased with the number of epochs and eventually converged to 100 %, while the accuracy of the validation set could be well stabilized at about 100 % after 5000 epochs. Finally, the accuracy of the MLP model on the test set was close to 100 %. According to our findings, the MLP neural network and fluorescence intensity have great potential applications in identifying the authenticity of honey.

Geographical origin discrimination of Chenpi using machine learning and enhanced mid-level data fusion

Chenpi, or dried tangerine peel, is a traditional Chinese ingredient valued in medicine and edible for its digestive and respiratory benefits. The geographical origin of Chenpi is important, as it can impact its quality, active compounds and market value. This study develops a strategy to distinguish Chenpi samples on its origin. Thirty-nine samples from eight regions in Xinhui district (Guangdong, China) are analyzed by gas chromatography (GC) and mid-infrared (MIR) technique. Four machine learning methods are employed to establish discrimination models based on GC and MIR data, with two mid-level data fusion strategies to combine the data. The results show that data fusion significantly improves Chenpi origin discrimination. The K-nearest neighbors and artificial neural network models, using modified mid-level data fusion, provide the best performance, misclassified only one sample. Machine learning in combination with modified mid-level data fusion strategy provides effective classification of Chenpi samples from different geographical origins.

Recent advances of machine learning in the geographical origin traceability of food and agro-products: A review

The geographical origin traceability of food and agro-products has been attracted worldwide. Especially with the rise of machine learning (ML) technology, it provides cutting-edge solutions to erstwhile intractable issues to identify the origin of food and agro-products. By utilizing advanced algorithms, ML can extract feature information of food and agro-products that is closely related to origin and, more accurately, identify and trace their origins, which is of great significance to the entire food and agriculture industry. This paper provides a comprehensive overview of the state-of-the-art applications of ML in the geographical origin traceability of food and agro-products. First, commonly used ML methods are summarized. The paper then outlines the whole process of preparation for modeling, model training as well as model evaluation for building traceability models-based ML. Finally, recent applications of ML combined with different traceability techniques in the field of food and agro-products are revisited. Although ML has made many achievements in solving the geographical origin traceability problem of food and agro-products, it still has great development potential. For example, the application of ML is yet insufficient in the geographical origin traceability using DNA or computer vision techniques. The ability of ML to predict the geographical origin of food and agro-products can be further improved, for example, by increasing model interpretability, incorporating data fusion strategies, and others.

Vibrational spectroscopy data fusion for enhanced classification of different milk types

The aim of this study is to classify seven types of Irish milk (butter, fresh, heart active, lactose free, light, protein, and slimline), supplied by a specific company, using vibrational spectroscopy methods: Near infrared (NIR), mid infrared (MIR), and Raman spectroscopy. In this regard, chemometric methods were used, and the impact of spectral data fusion on prediction accuracy was evaluated. A total of 105 samples were tested, with 21 used in the test set. The study assessed principal component analysis (PCA), partial least square discriminant analysis (PLS-DA), and sequential and orthogonalized partial least squares linear discriminant analysis (SO-PLS-LDA) for classifying different milk types. The prediction accuracy, when applying PLS-DA on individual blocks of data and low-level fused data, did not exceed 85.71 %. However, implementing the SO-PLS-LDA strategy significantly improved the accuracy to 95 %, suggesting a promising method for the development of classification models for milk using data fusion strategies.

Detection of physical descaling damage in carp based on hyperspectral images and dimension reduction of principal component analysis combined with pixel values

The surface of carp is easily damaged during the descaling process, which jeopardizes the quality and safety of carp products. Damage recognition realized by manual detection is an important factor restricting the automation in the pretreatment. For the commonly used methods of mechanical and water-jet descaling, damage area recognition according to the hyperspectral data was proposed. Two discrimination models, including decision tree (DT) and self-organizing feature mapping (SOM), were established to recognize the damaged and normal descaling area with the average spectral value. The damage-discrimination model based on DT was determined to be the optimal one, which possessed the best model performance (accuracy = 96.7%, sensitivity = 96.7%, specificity = 96.7%, F1-score = 96.7%). Considering the efficiency and precision of damage-area recognition and visualization, the principal component analysis (PCA) combined with pixel values statistical analysis was used to reduce the dimension of hyperspectral images at the image level. Through statistical analysis, the value 0 was used as the threshold to distinguish the normal area and the damaged area in the PC image to achieve preliminary segmentation. Then, the spectral values of the initially discriminated damage area were input into the DT discrimination model to realize the final discriminant of damaged area. On this basis, the position information of the damaged area could be used to realize the visualization. The final visualization maps for mechanical and water-jet descaling damage were obtained by image morphology processing. The average recognition accuracy can reach 94.9% and 90.3%, respectively. The results revealed that the hyperspectral imaging technique has great potential to recognize the carp damage area nondestructively and accurately under descaling processing. PRACTICAL APPLICATION: This study demonstrated that hyperspectral imaging technique can realize the carp damage area detection nondestructively and accurately under descaling processing. With the advantages of nondestructive and rapid, hyperspectral imaging system and the method can be widely expanded and applied to the quality detection of other freshwater fish pretreatment.

Handheld Devices for Food Authentication and Their Applications: A Review

This review summarises miniaturised technologies, commercially available devices, and device applications for food authentication or measurement of features that could potentially be used for authentication. We first focus on the handheld technologies and their generic characteristics: (1) technology types available, (2) their design and mode of operation, and (3) data handling and output systems. Subsequently, applications are reviewed according to commodity type for products of animal and plant origin. The 150 applications of commercial, handheld devices involve a large variety of technologies, such as various types of spectroscopy, imaging, and sensor arrays. The majority of applications, ~60%, aim at food products of plant origin. The technologies are not specifically aimed at certain commodities or product features, and no single technology can be applied for authentication of all commodities. Nevertheless, many useful applications have been developed for many food commodities. However, the use of these applications in practice is still in its infancy. This is largely because for each single application, new spectral databases need to be built and maintained. Therefore, apart from developing applications, a focus on sharing and re-use of data and calibration transfers is pivotal to remove this bottleneck and to increase the implementation of these technologies in practice.

Application of E-nose technology combined with artificial neural network to predict total bacterial count in milk

Total bacterial count (TBC) is a widely accepted index for assessing microbial quality of milk, and cultivation-based methods are commonly used as standard methods for its measurement. However, these methods are laborious and time-consuming. This study proposes a method combining E-nose technology and artificial neural network for rapid prediction of TBC in milk. The qualitative model generated an accuracy rate of 100% when identifying milk samples with high, medium, or low levels of TBC, on both the testing and validating subsets. Predicted TBC values generated by the quantitative model demonstrated strong coefficient of multiple determination (R² > 0.99) with reference values. Mean relative difference between predicted and reference values (mean ± standard deviation) of TBC were 1.1 ± 1.7% and 0.4 ± 0.8% on the testing and validating subsets involving 24 and 28 tested samples, respectively. Paired t-test implied that the difference between predicted and reference values of TBC was insignificant for both the testing and validating subsets. As low as ∼1 log cfu/mL of TBC present in tested samples were precisely predicted. Results of this study indicated that combination of E-nose technology and artificial neural network generated reliable predictions of TBC in milk. The method proposed in this study was reliable, rapid, and cost efficient for assessing microbial quality milk, and thus would potentially have realistic application in dairy section.

Estimating in vitro ruminal ammonia-N using multiple linear models and artificial neural networks based on the CNCPS nitrogenous fractions of cattle rations with low concentrate/roughage ratios

The objectives of this study were to investigate the relationship between the in vitro ruminal ammonia nitrogen (NH₃ -N) concentration and the Cornell Net Carbohydrate and Protein System (CNCPS) N-fractions of feeds for cattle and further compare the performance of developing multiple linear regression (MLR) and artificial neural network (ANN) models in estimating the NH₃ -N concentration in rumen fermentation. Two data sets were established, of which the training data set containing forty-five rations for cattle with concentrate/roughage ratios of 50:50, 40:60, 30:70, 20:80 and 10:90 used for developing models and the test data set containing ten other rations with the same concentrate/roughage ratios with the training data set were used for validating of models. The NH₃ -N concentrations of feed samples were measured using an in vitro incubation technique. The CNCPS N-fractions (g), for example PB₁ (rapidly degraded true protein), PB₂ (neutral detergent soluble nitrogen), PB₃ (acid detergent soluble nitrogen) of rations, were calculated based on chemical analysis. Statistical analysis indicated that the NH₃ -N concentration (mg) was significantly correlated with the CNCPS N-fractions (g) PB₁ , PB₂ and PB₃ in a multiple linear pattern: NH₃ -N = (130.70±33.80) PB₁  + (155.83±17.89) PB₂ - (85.44±37.69) PB₃  + (42.43±1.05), R²  = 0.77, p < 0.0001, n = 45. The results indicated that both MLR and ANN models were suitable for predicting in vitro NH₃ -N concentration of rations using CNCPS N-fractions PB₁ , PB₂ , and PB₃ as independent variables while the neural network model showed better performance in terms of greater r² , CCC and lower RMSPE between the observed and predicted values.

The Use of Artificial Neural Networks and a General Discriminant Analysis for Predicting Culling Reasons in Holstein-Friesian Cows Based on First-Lactation Performance Records

The aim of the present study was to verify whether artificial neural networks (ANN) may be an effective tool for predicting the culling reasons in cows based on routinely collected first-lactation records. Data on Holstein-Friesian cows culled in Poland between 2017 and 2018 were used in the present study. A general discriminant analysis (GDA) was applied as a reference method for ANN. Considering all predictive performance measures, ANN were the most effective in predicting the culling of cows due to old age (99.76-99.88% of correctly classified cases). In addition, a very high correct classification rate (99.24-99.98%) was obtained for culling the animals due to reproductive problems. It is significant because infertility is one of the conditions that are the most difficult to eliminate in dairy herds. The correct classification rate for individual culling reasons obtained with GDA (0.00-97.63%) was, in general, lower than that for multilayer perceptrons (MLP). The obtained results indicated that, in order to effectively predict the previously mentioned culling reasons, the following first-lactation parameters should be used: calving age, calving difficulty, and the characteristics of the lactation curve based on Wood's model parameters.

Advances in infrared spectroscopy combined with artificial neural network for the authentication and traceability of food

The authentication and traceability of food attract more attention due to the increasing consumer awareness regarding nutrition and health, being a new hotspot of food science. Infrared spectroscopy (IRS) combined with shallow neural network has been widely proven to be an effective food analysis technology. As an advanced deep learning technology, deep neural network has also been explored to analyze and solve food-related IRS problems in recent years. The present review begins with brief introductions to IRS and artificial neural network (ANN), including shallow neural network and deep neural network. More notably, it emphasizes the comprehensive overview of the advances of the technology combined IRS with ANN for the authentication and traceability of food, based on relevant literature from 2014 to early 2020. In detail, the types of IRS and ANN, modeling processes, experimental results, and model comparisons in related studies are described to set forth the usage and performance of the combined technology for food analysis. The combined technology shows excellent ability to authenticate food quality and safety, involving chemical components, freshness, microorganisms, damages, toxic substances, and adulteration. As well, it shows excellent performance in the traceability of food variety and origin. The advantages, current limitations, and future trends of the combined technology are further discussed to provide a thoughtful viewpoint on the challenges and expectations of online applications for the authentication and traceability of food.

Application of machine learning to improve dairy farm management: A systematic literature review

In recent years, several researchers and practitioners applied machine learning algorithms in the dairy farm context and discussed several solutions to predict various variables of interest, most of which were related to incipient diseases. The objective of this article is to identify, assess, and synthesize the papers that discuss the application of machine learning in the dairy farm management context. Using a systematic literature review (SLR) protocol, we retrieved 427 papers, of which 38 papers were determined as primary studies and thus were analysed in detail. More than half of the papers (55 %) addressed disease detection. The other two categories of problems addressed were milk production and milk quality. Seventy-one independent variables were identified and grouped into seven categories. The two prominent categories that were used in more than half of the papers were milking parameters and milk properties. The other categories of independent variables were milk content, pregnancy/calving information, cow characteristics, lactation, and farm characteristics. Twenty-three algorithms were identified, which we grouped into four categories. Decision tree-based algorithms are by far the most used followed by artificial neural network-based algorithms. Regression-based algorithms and other algorithms that do not belong to the previous categories were used in 13 papers. Twenty-three evaluation parameters were identified of which 7 were used 3 or more times. The three evaluation parameters that were used by more than half of the papers are sensitivity, specificity, RMSE. The challenges most encountered were feature selection and unbalanced data and together with problem size, overfitting/estimating, and parameter tuning account for three-quarters of the challenges identified. To the best of our knowledge, this is the first SLR study on the use of machine learning to improve dairy farm management, and to this end, this study will be valuable not only for researchers but also practitioners in dairy farms.

Rapid detection of carmine in black tea with spectrophotometry coupled predictive modelling

Carmine is an artificial colorant commonly used by fraudulent food business participants in black tea adulteration, for purpose of gaining illegal profits. This study combined spectrophotometry with machine learning for rapid detection of carmine in black tea based on the spectral characteristics of tea infusion. The qualitative model demonstrated an accuracy rate of 100% for successful identification of the presence/absence of carmine in black tea. For quantitative analysis, the R² between carmine concentrations generated according to spectral characteristics and those determined with HPLC was 0.988 and 0.972, respectively, for black tea samples involved in the test subset and an independent dataset II. Paired t-test indicated that the difference was statistically insignificant (P values of 0.26 and 0.44, respectively). The method established in this study was rapid and reliable for detecting carmine in black tea, and thus could be used as a useful tool to identify black tea adulteration in market.

High-Throughput Phenotyping Approach for Screening Major Carotenoids of Tomato by Handheld Raman Spectroscopy Using Chemometric Methods

Our objective was to develop a rapid technique for the non-invasive profiling and quantification of major tomato carotenoids using handheld Raman spectroscopy combined with pattern recognition techniques. A total of 106 samples with varying carotenoid profiles were provided by the Ohio State University Tomato Breeding and Genetics program and Lipman Family Farms (Naples, FL, USA). Non-destructive measurement from the surface of tomatoes was performed by a handheld Raman spectrometer equipped with a 1064 nm excitation laser, and data analysis was performed using soft independent modelling of class analogy (SIMCA)), artificial neural network (ANN), and partial least squares regression (PLSR) for classification and quantification purposes. High-performance liquid chromatography (HPLC) and UV/visible spectrophotometry were used for profiling and quantification of major carotenoids. Seven groups were identified based on their carotenoid profile, and supervised classification by SIMCA and ANN clustered samples with 93% and 100% accuracy based on a validation test data, respectively. All-trans-lycopene and β-carotene levels were measured with a UV-visible spectrophotometer, and prediction models were developed using PLSR and ANN. Regression models developed with Raman spectra provided excellent prediction performance by ANN (r_pre = 0.9, SEP = 1.1 mg/100 g) and PLSR (r_pre = 0.87, SEP = 2.4 mg/100 g) for non-invasive determination of all-trans-lycopene in fruits. Although the number of samples were limited for β-carotene quantification, PLSR modeling showed promising results (r_cv = 0.99, SECV = 0.28 mg/100 g). Non-destructive evaluation of tomato carotenoids can be useful for tomato breeders as a simple and rapid tool for developing new varieties with novel profiles and for separating orange varieties with distinct carotenoids (high in β-carotene and high in cis-lycopene).

Predicting the redshift on the ultraviolet spectrum using the peak area method

The ultraviolet (UV) absorption peaks of NaCl, NaOH, and $\beta $β-phenylethylamine (PEA) in an aqueous solution move toward redshift. We proposed the peak area method for the quantitative analysis of PEA, NaCl, and NaOH. First, we obtained the predictable regularities of the redshift of the single component sample. Then, we obtained the regularities of the redshift of the UV spectrum for the mixture by the peak height and peak area methods. Finally, the BP-ANN algorithm was applied to determine the concentration of the mixture using the peak height and peak area method. The results of the testing set showed that correlation coefficients (${{\rm R}^2}$R²) of 0.992, 0.993, and 0.992 were obtained by peak height method and 0.997, 0.998, and 0.998 were obtained by peak area method for NaCl, NaOH, and PEA, respectively. Meanwhile, the relative errors of the prediction of NaCl, NaOH, and PEA obtained by peak area method were less than 3%, whereas the REP of NaCl, NaOH, and PEA obtained by peak height method were more than 5%. Compared with the results of the peak height method, the results showed that the peak area was more accurate than the peak height in predicting the redshift of the UV spectrum.