Near-infrared technology in agriculture: Rapid, simultaneous, and non-destructive determination of inner quality parameters on intact coffee beans

Agus Arip Munawar; Kusumiyati Kusumiyati; Andasuryani Andasuryani; Yusmanizar Yusmanizar; Adrizal Adrizal

doi:10.1515/opag-2022-0290

Article Open Access

Near-infrared technology in agriculture: Rapid, simultaneous, and non-destructive determination of inner quality parameters on intact coffee beans

Agus Arip Munawar , Kusumiyati Kusumiyati , Andasuryani Andasuryani , Yusmanizar Yusmanizar and Adrizal Adrizal

Published/Copyright: June 18, 2024

Published by

Become an author with De Gruyter Brill

Submit Manuscript Author Information Explore this Subject

From the journal Open Agriculture Volume 9 Issue 1

Abstract

This study delves into the ability of near infrared (NIR) techniques by means of a self-developed portable sensing device near-infrared reflectance spectroscopy (NIRS) i16 USK instrument to accurately predict the moisture content (MC) and chlorogenic acid (CGA) within intact coffee beans through the development of calibration models. Spectral absorbance measurements were conducted across the 1,000‒2,500 nm wavelength range. Leveraging two multivariate calibration approaches namely principal component regression and partial least square regression (PLSR) for 74 bulk coffee beans (60 g) in calibration and 36 bulk coffee beans samples in external validation. The results reveal a notably high determination coefficient (R ²) of 0.984 for MC and 0.908 for CGA in calibration using PLSR, indicating the feasibility of rapid, simultaneous, and non-destructive prediction. Furthermore, upon external validation, the PLSR model exhibited consistent predictive performance, with R ² values for MC and CGA contents reaching 0.978 and 0.846, respectively. Consequently, these outcomes underscore NIR as an effective, concurrent, and non-invasive means to assess the quality parameters and attributes of intact coffee beans, presenting promising prospects for the advancement of coffee quality evaluation.

Keywords: NIRS; coffee; agriculture; technology

1 Introduction

Coffee is a globally significant agricultural commodity and a beloved drink for many people due to its unique taste, aroma, and nutritional benefits. It has various uses, from being a daily beverage to a food additive or ingredient in other edible products. It is essential for consumers to receive high-quality coffee at a reasonable cost, emphasizing the need for rigorous quality control in the coffee sector [1]. In addition, coffee waste like spent coffee grounds can serve as an effective soil fertilizer and enhancer, owing to their rich content of carbon (C), nitrogen (N), phosphorus (P), and potassium (K) [2]. The internal qualities that predominantly influence the final quality of coffee include moisture levels and chlorogenic acid (CGA). The CGA is recognized for its health benefits and its impact on coffee’s flavor, acidity, and bitterness [3]. Moisture content (MC) is a key quality metric for raw coffee beans, with importing nations having specific standards. An MC below 8% can lead to unappealing appearances in coffee beans, whereas above 12.5% can encourage the growth of fungi and mycotoxins [4].

Monitoring and controlling the MC of coffee beans is crucial for ensuring product safety, quality, and longevity. Beans with elevated moisture levels are prone to mold and mycotoxin development, posing significant health risks. Therefore, maintaining optimal MC is essential for preserving the integrity and shelf life of coffee products. On the other hand, evaluating the CGA content in coffee beans provides valuable insights into both the flavor profile and potential health benefits of the final coffee product. CGA, as a group of antioxidants, contributes to the flavor, acidity, and overall sensory experience of the coffee, while also potentially offering health-promoting properties. As such, assessing both the MC and CGA levels in coffee beans plays a vital role in ensuring quality, flavor, and potential health attributes of the end product.

Importing countries often have strict regulatory standards in place for the MC and CGA levels in coffee beans to ensure product safety and quality. The regulations set limits for the permissible MC to prevent mold growth and mycotoxin formation, safeguarding consumers from potential health hazards. Additionally, these standards often specify acceptable ranges for the CGA, influencing the flavor profile, potential health benefits, and overall quality of the coffee. By enforcing these regulatory standards, importing countries aim to maintain the safety, consistency, and quality of the coffee products available to consumers, ensuring that they meet specific criteria for MC, CGA levels, and overall product quality.

Most common methods to assess coffee quality often rely on solvent extraction and other laboratory processes. These techniques are typically slow, destructive, labor-intensive, and might even have environmental concerns, making them less ideal for an industry that demands quick and straightforward quality checks [5,6,7]. Recently, there has been a shift towards finding more efficient alternatives for evaluating food and agricultural product quality. These new methods aim to be rapid, straightforward, involve fewer chemicals, and be non-destructive.

One such technique is near-infrared reflectance spectroscopy (NIRS), an eco-friendly and dependable method for examining agricultural products. NIRS operates on the principle of how near-infrared radiation interacts with a biological entity. This radiation spans the 780‒2,500 nm range of the electromagnetic spectrum [8,9]. When subjected to this radiation, an object’s spectral properties change based on its chemical properties. The foundational principle of NIRS lies in its ability to probe the interactions of near-infrared radiation with biological and organic materials. When near-infrared (NIR) light is shone on a sample, specific wavelengths are absorbed based on the sample’s molecular composition, while others are reflected [10]. By analyzing the reflected light spectrum, it is possible to derive insights into the molecular composition of the sample. Given that moisture and CGA have unique molecular structures, their presence and concentration in coffee beans cause distinct spectral signatures that can be detected and quantified using NIRS.

Numerous research articles highlight the utility of NIRS in assessing the quality of different products, from fruits [11,12,13], meat and dairy products [14,15,16], biodiesel [17,18], to soils [19,20,21]. The growing adoption of NIRS by many industries and its frequent mention in scientific literature underline its significance. Given this context, our focus is to explore the feasibility of using NIRS to determine quality parameters in intact coffee beans, specifically moisture and CGA concentrations. A novel aspect of this work is the employment of a self-developed, portable NIRS instrument, namely portable sensing device (PSD) NIRS i16 USK, which is designed for real-time, on-site analysis. Furthermore, our study is particularly focused on green coffee beans of both Arabica and Robusta varieties, sourced from various regions in Indonesia. This targeted examination of Indonesian origins allows for a detailed understanding of the geographical influence on bean characteristics.

2 Materials and methods

2.1 Samples

In this study, analysis was conducted on a total of 110 comprehensive batches, with each batch comprising 60 g of non-roasted intact coffee beans sourced from four distinct Indonesian regions, namely, Gayo Aceh, West Java, West Sumatera, and West Nusa Tenggara. A diverse geographic sample was represented by these batches. Both Arabica and Robusta coffee varieties were included in the sample collection, providing a comprehensive representation of the different coffee types found within these regions. These samples were obtained and harvested selectively from local farmers in those regions in Indonesia. To obtain varied MC and CGA compositions, coffee bean samples were selectively handpicked with a degree of maturity from unripe to ripe maturity. Samples were split into two group datasets, namely calibration, used to develop MC and CGA prediction models, and validation dataset used to validate and test the models' performances. For NIRS analysis and MC measurement, the sample consists of fully intact coffee beans. This intact bean sample configuration ensures that the NIRS analysis provides representative and accurate measurements of the MC within the entire beans.

On the other hand, for CGA determination, the sample is prepared in powdered form. The grinding of the coffee beans into a fine powder allows for enhanced surface area exposure, aiding in the precise measurement of the CGA through NIRS analysis. To ensure stabilization, the coffee bean batches were maintained at a room temperature of 26°C for a period of 3 days prior to their NIR spectral evaluation and subsequent analyses. For each assessment, a quantity ranging from 50 to 60 g of coffee beans was placed into a vial sample holder.

2.2 NIR spectral data acquisition

The NIR spectral data for intact coffee beans were acquired and collected using the PSD NIRS i16 USK Instrument, which is equipped with a laser for the light source as illustrated in Figure 1. Each coffee sample was measured across a high-resolution wavelength range from 1,000 to 2,500 nm [22]. The obtained data, which averaged 32 scans for each sample, were subsequently archived on a dedicated computer system.

Figure 1

Spectra data acquisition of intact coffee beans in NIR wavelength region.

The instrument shines near-infrared light on the coffee beans. This light penetrates the beans and is reflected back. The molecular vibrations caused by specific chemical bonds in the coffee beans absorb certain wavelengths of the NIR light. The detector in the instrument captures the reflected light and produces a spectrum, which is a plot of light intensity.

2.3 MC measurement

To measure the MC of coffee beans in accordance with ISO 6673 [23], a consistent approach for determining MC is vital for assessing the quality and managing the storage of coffee beans. Approximately 5 g of the prepared sample is weighed accurately and placed in a moisture dish. The sample is then dried in an oven set at 103°C ± 2°C. The drying process continues until a constant weight is achieved, which typically occurs after 4‒6 h [4,7]. After the drying period, the sample is removed from the oven and allowed to cool in a desiccator to prevent moisture absorption from the atmosphere. Once cooled, the sample is re-weighed. The difference in weight before and after the drying process indicates the amount of moisture lost. This difference is used to calculate the MC of the coffee sample, expressed as a percentage.

2.4 CGA measurement

To determine the content of CGAs in coffee beans via HPLC, the beans first undergo an extraction process. Typically, this involves grinding the beans to a fine powder, allowing for effective solvent penetration. A known weight of the coffee powder is then mixed with a solvent, commonly a mixture of water, methanol, or acetonitrile, sometimes acidified with dilute acid. This mixture is sonicated or agitated for a set duration to facilitate the extraction of CGA from the coffee matrix. After extraction, the solution is filtered to remove any solid residues, and the filtrate, containing the extracted CGA, is collected. The filtrate might also be subjected to further purification, like centrifugation, to ensure its clarity. An aliquot of this solution is then injected into the HPLC system equipped with a suitable column, often a reversed-phase C18 column.

The system separates the CGA based on their respective interactions with the column material and their solubility in the mobile phase. The eluted CGA compounds are detected, typically using a UV or PDA detector, at wavelengths where they exhibit strong absorbance, commonly around 320–325 nm. The area under the peaks corresponding to CGA compounds, when compared to a standard curve created with known concentrations of CGA standards, gives the concentration of CGA in the coffee sample. Finally, the results are expressed as milligrams of CGA per gram or per kilogram of coffee [5].

2.5 MC and CGA prediction by NIRS

The NIRS offers a rapid, non-destructive, and simultaneous technique for predicting various parameters in agricultural products, including the MC and CGAs of coffee beans. Calibration is an essential aspect of NIRS, ensuring the accuracy and reliability of predictions. For NIRS to be effectively utilized in predicting the MC and CGA levels in coffee beans, robust calibration models were established. The calibration process begins with the selection of a diverse and representative set of coffee bean samples. These samples cover the expected range of moisture and CGA contents as presented in Table 1.

Table 1

Descriptive statistics of actual moisture and CGA contents

Statistical parameters	CGA (%)	Water content (%)
# n Sample	74	74
Mean	8.781	0.216
Max	12.478	0.315
Min	6.490	0.159
Range	5.988	0.155
Std. Deviation	1.454	0.043
Variance	2.114	0.001
RMS	8.899	0.220
Median	8.587	0.210

The reference data and the corresponding spectra are then combined to develop a calibration model using multivariate statistical methods namely PLSR and principal component regression (PCR). The quality of the calibration model is assessed through cross-validation, wherein a subset of samples is predicted by the model, and the NIRS predictions are compared to the actual reference moisture and CGA values. Goodness of fit and prediction in the form of coefficient R ², prediction error RMSE, and ratio prediction to deviation (RPD) indexes determine the accuracy of the calibration and validation [8,24].

3 Results and discussion

A typical spectrum of intact coffee beans in the NIR wavelength region is presented in Figure 2. It provides information about their chemical and physical properties. When coffee beans are irradiated with NIR light, various chemical constituents within the beans absorb specific wavelengths, resulting in a unique spectral fingerprint. Around 1,450 and 1,930 nm, the absorption peaks are mainly due to the O–H stretching vibrations of water molecules. The intensity of these bands can be related to the MC of the coffee beans. Changes in the intensity and shape of these peaks can reflect fluctuations in water content [25]. This is essential as the moisture level affects the quality, flavor, and shelf-life of coffee beans. Moreover, absorption bands around 1,720, 2,300, and 2,340 nm are attributed to the C–H stretching vibrations from lipids and cellulose in the beans. The intensity of these bands can indicate the fat or lipid content in the coffee beans.

Figure 2

Typical NIR spectrum of intact coffee bean.

Vibration bands occurring around 2,100 and 2,300 nm can be attributed to the N–H stretching vibrations from proteins. It plays a role in the texture and flavor of brewed coffee. The bands around 2,100 and 2,300 nm correspond to N–H stretching vibrations in proteins. An analysis of these bands can be used to derive the protein content and, in some cases, hint at the bean’s overall maturity and quality. On the other hand, CGA is essential for flavor development during roasting and plays a role in the health benefits of coffee. While these compounds do exhibit absorption in the NIR region, the exact bands can overlap with other components. Therefore, precise quantification of CGA using NIR often requires sophisticated statistical analysis methods and a well-developed calibration set.

3.1 MC prediction

NIR spectroscopy is a non-destructive analytical technique that utilizes the near-infrared region of the electromagnetic spectrum to determine the chemical composition of a sample. In the context of coffee beans, NIRS can be used to predict MC, which is crucial for assessing the quality and shelf life of the beans. In this study, PCR and PLSR were employed to develop prediction models used to determine MC as presented in Table 2. Raw spectral data of intact coffee beans were used as independent variables for PCR and PLSR. PLSR seems to be slightly more accurate in predicting water content with higher correlation and determination coefficients than PCR. Nevertheless, both methods provided good prediction performance to determine MC with correlation R ² of 0.979 for PCR and 0.984 for PLSR, respectively. A scatter plot derived from actual and predicted MC by PCR and PLSR is presented in Figure 3.

Table 2

MC prediction by means of PCR and PLSR

Method	Statistical indicators
Method	R ²	r	RMSE	RPD
PCR	0.979	0.989	0.006	7.217
PLSR	0.984	0.992	0.005	8.661

PCR, principal component regression; PLSR, partial least squares regression; R ², coefficient of determination; r, correlation coefficient; RMSE, root mean square error; RPD, ratio prediction to deviation.

Figure 3

Prediction performance of (a) PCR and (b) PLSR to determine MC.

The PCR and PLSR are two common multivariate calibration methods used in conjunction with NIRS to model the relationship between the spectral data and the property of interest. When comparing PCR and PLSR for predicting the MC in coffee beans using NIRS spectral data, it is essential to consider various aspects, including the R-square value, model complexity, predictive performance, variable selection, computational efficiency, and practical implications. Both models exhibit high R-square values, 0.979 for PCR and 0.984 for PLSR, suggesting a strong fit to the data with 97.9 and 98.4% of the variability in MC being explained by each model respectively. However, the slightly higher R-square value for PLSR indicates a marginally better fit to the data, which can be crucial for precision in predictive analytics.

Both PCR and PLSR models in this instance utilized seven factors, showcasing a balance between model complexity and the risk of overfitting. PCR, being less complex, might be more computationally efficient, particularly for extensive datasets. However, its simplicity comes with a potential risk of overfitting, especially when principal components do not capture the variability in the dependent variable efficiently. PLSR, on the other hand, is proficient in handling multi-collinearity and tends to offer more robust models by focusing on components that are most relevant to the dependent variable, thereby often achieving better predictive performance on external datasets.

Another pivotal point of distinction lies in variable selection. PCR inherently does not perform variable selection and utilizes all variables or principal components, possibly incorporating irrelevant or noisy variables. In contrast, PLSR, by design, performs variable selection, emphasizing variables that share a strong relationship with the response variable, which can enhance model interpretability and accuracy.

In terms of practical implications, the simplicity and computational efficiency of PCR might make it a preferable choice in scenarios where resources are constrained and a slight compromise in accuracy is acceptable. However, when the highest level of accuracy is sought and computational resources are ample, PLSR emerges as a more favorable option owing to its superior predictive accuracy and robustness. In brief, while both PCR and PLSR have demonstrated commendable predictive abilities in determining MC using NIRS spectral data, the choice between them should be meticulously made, considering the specific requirements and constraints of the application. The slight edge PLSR has in terms of accuracy, robustness, and variable selection needs to be weighed against the computational simplicity and efficiency of PCR, especially in practical, resource-limited scenarios.

3.2 CGA content prediction

The prediction performance of NIR spectroscopy, as employed in the assessment of CGA content in intact coffee beans, is also a discussion point in the field of analytical chemistry. Particularly, this discourse is inclined towards understanding the comparison between PCR and PLSR methods of prediction. In this case, PLSR also offers an R ² value of 0.908, which is significantly higher compared to the 0.774 value provided by the PCR method as presented in Table 3 and Figure 4. Consequently, these data tend to suggest a stronger predictive performance by PLSR in the prediction of CGA content.

Table 3

CGA prediction by means of PCR and PLSR

Method	Statistical indicators
Method	R ²	r	RMSE	RPD
PCR	0.774	0.879	0.618	2.265
PLSR	0.908	0.952	0.435	3.209

Figure 4

Prediction performance of (a) PCR and (b) PLSR to determine CGA content.

Indeed, the RPD values seem to confirm this presumption. RPD is a useful statistic for assessing the quality or reliability of a calibration model, signifying that the model can be used for quantitative predictions with high confidence. Here, the RPD value for PLSR stands at 3.209, indicating a strong prediction model. Conversely, PCR’s RPD score was 2.265, falling into the sufficient performance category, suggesting that it might be less accurate for prediction tasks compared to the PLSR.

However, it is crucial not to take these numbers at face value, solely. Although PLSR attains better R ² and RPD values in this scenario, the appropriateness, efficiency, and accuracy of PCR and PLSR might vary significantly under different contexts. The prediction capacity is influenced by factors such as the complexity and nature of the prediction task, the quality, dimensionality, and abundance of the sampling data, and the accuracy of implementing the models. The additional information about the root mean squared error of calibration (RMSEC) for PCR being 0.618 and for PLSR being 0.435 adds further insights into the predictive capabilities of the two methods. RMSEC is a measure of the average deviation between the measured and predicted values in the calibration set, providing an assessment of the accuracy of the calibration models. Comparing the RMSEC values, we can observe that PLSR yields a significantly lower value of 0.435 compared to PCR’s value of 0.618. A lower RMSEC indicates that the predicted CGA from the PLSR model is closer to the actual observed values. This enhances the confidence in the predictive performance of PLSR compared to PCR. Hence, it can be inferred that PLSR demonstrates superior performance in predicting the CGA content of intact coffee beans using NIR spectroscopy. The higher R ² value (0.908), lower RMSEC (0.435), and higher RPD (3.41) for PLSR compared to PCR indicate greater accuracy, better prediction, and higher reliability.

Nevertheless, it is important to note that these findings are specific to the given data and should be interpreted cautiously. The predictive performance of PCR and PLSR can vary depending on factors such as the quality and representativeness of the calibration dataset, the specific spectral characteristics of the samples, and potential model overfitting or underfitting. Therefore, further validation and analysis, including external validation and comparison with other models, would be beneficial in obtaining a comprehensive understanding of the performance prediction of NIR spectroscopy for CGA in intact coffee beans.

3.3 Model validation

The application of near-infrared spectroscopy using partial least squares regression (NIRS-PLSR) in predicting moisture and CGA content on intact coffee beans reveals some important findings. Another important aspect to consider when evaluating the predictive performance of models is their ability to generalize to new and independent datasets. This can be examined through validation techniques. The model was tested using 36 external coffee bean samples from the validation dataset to simultaneously determine MC and CGA by means of NIR spectral data. Chosen models based on the PLSR approach were used in validation since it provided better prediction performance than PCR during calibration. The validation performance of the NIRS-PLSR model for rapidly assessing moisture and CGA contents is presented in Figure 5.

Figure 5

Prediction performance of PLSR-NIRS to determine (a) MC and (b) CGA contents.

Analyzing the data for MC prediction, we observe a high determination coefficient (R ²) of 0.978. This indicates a remarkable fit for the prediction model, with approximately 97.8% of the variation in MC being sufficiently explained by the model. In other words, it demonstrates that the model can accurately predict MC with a high degree of confidence, which is paramount for controlling the quality of coffee beans. The root mean squared error of prediction (RMSEP) value is 0.007 for this prediction task, which is substantially low, indicating that deviation of predicted values from actual values is minimal. These statistics showcase the efficacy and reliability of the NIRS-PLSR model in the prediction of MC in coffee beans.

On the contrary, its performance in predicting the CGA content of coffee beans, while still noteworthy, shows a slightly lower degree of accuracy. The generated R ² value is 0.846, which still accounts for a substantial proportion of approximately 84.6% of variability in CGA content. Although lower than the R² for MC, this figure still denotes an acceptable degree of fit for prediction. However, the RMSEP for CGA prediction is substantially higher at 0.712. This value implies a greater discrepancy between predicted values and observed values compared to the MC prediction model. Elevated RMSEP often signals higher prediction errors, which suggests that the NIRS-PLSR model’s precision and accuracy in predicting CGA content might see some room for improvement.

The results indicate that the NIRS-PLSR model exhibits commendable predictive performance. In the context of the coffee industry, where monitoring the moisture and CGA content of beans is pivotal for quality control, NIRS-PLSR can be a valuable tool. This predictive performance combined with the non-destructive nature of NIR spectroscopy can potentially revolutionize the quality assessment procedures in the coffee industry. Yet, while the prediction performance for MC seems outstanding, the prediction for CGA, though still respectable, contains a notable margin for further enhancement. The future aim should be to refine the prediction model for CGA by considering other influencing factors or by optimizing the PLSR algorithm, thus ensuring a uniformly high level of accuracy for both moisture and CGA content predictions. It is through such efforts that we can fully harness and further explore the potential of NIRS-PLSR-based models in the coffee industry.

Our analysis revealed that the MC of coffee beans is a critical factor for regulatory compliance and preserving product quality. The tested samples remained within the optimal range for MC, ensuring they are less susceptible to fungal contamination, thereby conforming to the stringent food safety standards of the EU. This adherence to MC standards is indicative of the effectiveness of our preservation and transportation processes. Moreover, the CGAs content, as assessed by NIRS, correlated strongly with the anticipated flavor profiles, suggesting that our coffee beans possess the high-quality attributes favored by European consumers.

The CGA levels not only reflect the antioxidant potential of our product but also serve as a valuable indicator for gauging consumer preference, given that the EU market exhibits a sophisticated palate with a preference for nuanced flavor notes. Importantly, the data derived from these parameters have implications for product labeling, where accurate information regarding antioxidant content can be leveraged for market differentiation. The discussions thus elucidate how MC and CGA parameters intersect with key market considerations such as regulatory compliance, quality assurance, consumer preferences, and marketing strategies, offering insights into the broader implications of these quality indicators on the trade and consumption of imported coffee within the EU context.

4 Conclusion

The use of NIR spectroscopy, coupled with PLSR for prediction of key properties in coffee beans, demonstrates significant promise. This premise was explored through a comparative analysis of PLSR and PCR and also by an examination of the predictive capabilities of PLSR models tested on an external validation dataset. Comparatively, PLSR depicted superior capabilities than PCR, as evidenced by higher determination coefficient (R²) values and RPD index, and a lower RMSEC. For the prediction of CGA content, PLSR had an R ² value of 0.90 as compared to PCR’s 0.74, signifying a better fit for the model.

The predictive performance of the PLSR-based NIRS model was further tested using 36 external validation data sets to determine moisture and CGA in coffee beans. The results indicate high reliability and efficiency for MC prediction with an R² value of 0.978 and an RMSEP of 0.007. CGA content prediction also depicted a fairly good R² value of 0.846, highlighting the model’s ability to account for about 84.6% of variability in data; however, a higher RMSEP of 0.712 indicates a higher prediction error compared to the MC prediction model. Overall, these findings notify the robustness and effectiveness of NIRS combined with PLSR in predicting quality parameters like moisture and CGA content in coffee beans. Nevertheless, it also underlines the need for refining the CGA prediction model, thereby ensuring enhanced predictive accuracy across multiple key parameters. This assurance of predictive accuracy is key for the wider acceptance of NIRS-PLSR-based models for rapid, simultaneous, and non-destructive method in the coffee industry’s quality assessment procedures.

Acknowledgements

The authors are grateful for the reviewer’s valuable comments that improved the manuscript.

Funding information: This presented work is funded by the Directorate for Research and Community Service (LPPM) Universitas Syiah Kuala through Riset Kolaborasi Indonesia RKI scheme 2024.
Author contributions: All authors accepted the responsibility for the content of the manuscript and consented to its submission, reviewed all the results, and approved the final version of the manuscript. AAM designed the experiments and wrote the original manuscript, KK performed calibration modeling and validation, AA developed data analysis and graph interpretation, YY carried out laboratory analysis and prepared samples, and AA reviewed and revised the manuscript with contributions from all authors.
Conflict of interest: The authors state no conflict of interest.
Data availability statement: The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.

References

[1] Min J, Cao Z, Cui L, Li F, Lu Z, Hou Y, et al. The association between coffee consumption and risk of incident depression and anxiety: Exploring the benefits of moderate intake. Psychiatry Res. 2023;326(April):115307. 10.1016/j.psychres.2023.115307.Search in Google Scholar PubMed

[2] Tombarkiewicz B, Antonkiewicz J, Lis MW, Pawlak K, Trela M, Witkowicz R, et al. Chemical properties of the coffee grounds and poultry eggshells mixture in terms of soil improver. Sci Rep. 2022;12(1):1–10. 10.1038/s41598-022-06569-x.Search in Google Scholar PubMed PubMed Central

[3] Lemos MF, de Andrade Salustriano N, de Souza Costa MM, Lirio K, da Fonseca AF, Pacheco HP, et al. Chlorogenic acid and caffeine contents and anti-inflammatory and antioxidant activities of green beans of conilon and arabica coffees harvested with different degrees of maturation. J Saudi Chem Soc. 2022;26(3). 10.1016/j.jscs.2022.101467.Search in Google Scholar

[4] Adnan A, von Hörsten D, Pawelzik E, Mörlein D. Rapid prediction of moisture content in intact green coffee beans using near infrared spectroscopy. Foods. 2017;6(6):38–44. 10.3390/foods6050038.Search in Google Scholar PubMed PubMed Central

[5] Shan J, Suzuki T, Suhandy D, Ogawa Y, Kondo N. Chlorogenic acid (CGA) determination in roasted coffee beans by Near Infrared (NIR) spectroscopy. Eng Agric Env Food. 2014;7(4):139–42. 10.1016/j.eaef.2014.08.003.Search in Google Scholar

[6] Santos JR, Lopo M, Rangel AOSS, Lopes JA. Exploiting near infrared spectroscopy as an analytical tool for on-line monitoring of acidity during coffee roasting. Food Control. 2016;60:408–15. 10.1016/j.foodcont.2015.08.007.Search in Google Scholar

[7] Correia RM, Tosato F, Domingos E, Rodrigues RR, Aquino LF, Filgueiras PR, et al. Portable near infrared spectroscopy applied to quality control of Brazilian coffee. Talanta. 2018;176(August 2017):59–68. 10.1016/j.talanta.2017.08.009.Search in Google Scholar PubMed

[8] Pasquini C. Near infrared spectroscopy: A mature analytical technique with new perspectives – A review. Anal Chim Acta. 2018;1026:8–36. 10.1016/j.aca.2018.04.004.Search in Google Scholar PubMed

[9] de F, Pires C, Pereira RGFA, Baqueta MR, Valderrama P, Alves da Rocha R. Near-infrared spectroscopy and multivariate calibration as an alternative to the Agtron to predict roasting degrees in coffee beans and ground coffees. Food Chem. 2021;365(January):130471. 10.1016/j.foodchem.2021.130471.Search in Google Scholar PubMed

[10] Munawar AA, Meilina H, Pawelzik E. Near infrared spectroscopy as a fast and non-destructive technique for total acidity prediction of intact mango: Comparison among regression approaches. Comput Electron Agric. 2022;193(3):106657. 10.1016/j.compag.2021.106657.Search in Google Scholar

[11] Gabriëls SHEJ, Mishra P, Mensink MGJ, Spoelstra P, Woltering EJ. Non-destructive measurement of internal browning in mangoes using visible and near-infrared spectroscopy supported by artificial neural network analysis. Postharvest Biol Technol. 2020;166(April):111206. 10.1016/j.postharvbio.2020.111206.Search in Google Scholar

[12] Mishra P, Woltering E, El Harchioui N. Improved prediction of ‘Kent’ mango firmness during ripening by near-infrared spectroscopy supported by interval partial least square regression. Infrared Phys Technol. 2020;110(June):103459. 10.1016/j.infrared.2020.103459.Search in Google Scholar

[13] Hayati R, Munawar AA, Fachruddin F. Enhanced near infrared spectral data to improve prediction accuracy in determining quality parameters of intact mango. Data Br. Apr. 2020;30(1):112–6. 10.1016/j.dib.2020.105571.Search in Google Scholar PubMed PubMed Central

[14] López-Maestresalas A, Insausti K, Jarén C, Pérez-Roncal C, Urrutia O, Beriain MJ, et al. Detection of minced lamb and beef fraud using NIR spectroscopy. Food Control. 2019;98(October 2018):465–73. 10.1016/j.foodcont.2018.12.003.Search in Google Scholar

[15] Pieszczek L, Czarnik-Matusewicz H, Daszykowski M. Identification of ground meat species using near-infrared spectroscopy and class modeling techniques – Aspects of optimization and validation using a one-class classification model. Meat Sci. 2018;139(March 2017):15–24. 10.1016/j.meatsci.2018.01.009.Search in Google Scholar PubMed

[16] Grunert T, Stephan R, Ehling-Schulz M, Johler S. Fourier Transform Infrared Spectroscopy enables rapid differentiation of fresh and frozen/thawed chicken. Food Control. 2016;60:361–4. 10.1016/j.foodcont.2015.08.016.Search in Google Scholar

[17] García-martín JF, Alés-álvarez FJ, López-barrera MC, Martín-domínguez I, Álvarez-mateos P. Cetane number prediction of waste cooking oil-derived biodiesel prior to transesteri fi cation reaction using near infrared spectroscopy. Fuel. 2019;240(November 2018):10–5. 10.1016/j.fuel.2018.11.142.Search in Google Scholar

[18] Balabin RM, Lomakina EI, Safieva RZ. Neural network ( ANN) approach to biodiesel analysis: Analysis of biodiesel density, kinematic viscosity, methanol and water contents using near infrared (NIR) spectroscopy. Fuel. 2015;90(5):2007–15. 10.1016/j.fuel.2010.11.038.Search in Google Scholar

[19] dos Santos UJ, de Melo Dematte JA, Menezes RS, Dotto AC, Guimarães CC, Alves BJ, et al. Predicting carbon and nitrogen by visible near-infrared (Vis-NIR) and mid-infrared (MIR) spectroscopy in soils of Northeast Brazil. Geoderma Reg. 2020;23:154–61. 10.1016/j.geodrs.2020.e00333.Search in Google Scholar

[20] Zhang X, Sun W, Cen Y, Zhang L, Wang N. Predicting cadmium concentration in soils using laboratory and field reflectance spectroscopy. Sci Total Environ. 2019;650:321–34. 10.1016/j.scitotenv.2018.08.442.Search in Google Scholar PubMed

[21] Wang Q, Zhang H, Li F, Gu C, Qiao Y, Huang S. Assessment of calibration methods for nitrogen estimation in wet and dry soil samples with different wavelength ranges using near-infrared spectroscopy. Comput Electron Agric. 2021;186(December 2020):106181. 10.1016/j.compag.2021.106181.Search in Google Scholar

[22] Munawar AA, von Hörsten D, Wegener JK, Pawelzik E, Mörlein D. Rapid and non-destructive prediction of mango quality attributes using Fourier transform near infrared spectroscopy and chemometrics. Eng Agric Environ Food. Jul. 2016;9(3):208–15. 10.1016/j.eaef.2015.12.004.Search in Google Scholar

[23] Technical Committee ISO/TC 34. ISO 6673:2003 Green coffee Determination of loss in mass at 105 degrees C. Food products, Subcommittee SC 15, Coffee, 2003.Search in Google Scholar

[24] Nicolaï BM, Beullens K, Bobelyn E, Peirs A, Saeys W, Theron KI, et al. Nondestructive measurement of fruit and vegetable quality by means of NIR spectroscopy: A review. Postharvest Biol Technol. 2007;46(2):99–118. 10.1016/j.postharvbio.2007.06.024.Search in Google Scholar

[25] de Araújo TKL, Nóbrega RO, de Sousa Fernandes DD, de Araújo MCU, Diniz PHGD, da Silva EC. Non-destructive authentication of Gourmet ground roasted coffees using NIR spectroscopy and digital images. Food Chem. March, 2021;364:65–73. 10.1016/j.foodchem.2021.130452.Search in Google Scholar PubMed

Received: 2023-11-08

Revised: 2024-02-09

Accepted: 2024-04-13

Published Online: 2024-06-18

This work is licensed under the Creative Commons Attribution 4.0 International License.

Articles in the same Issue

https://doi.org/10.1515/opag-2022-0290

Keywords for this article

NIRS; coffee; agriculture; technology

Creative Commons

BY 4.0