Application | Quality Parameters | Data Analysis | Results | References |
---|---|---|---|---|
NIR-Hyperspectral imaging | Classification | PLS-DA, SVM | Prediction error: 3.8–23.1% (SVM) 4.4–34.4% (PLS-DA) | Cruz-Tirado et al. (2020) |
Hyperspectral imaging | Classification, FI | PLSR | 95% accuracy | Bayona et al. (2018) |
Hyperspectral imaging | FI, TP, AA | PLSR | R2 = 0.5 (RMSEP = 0.27, RPD = 1.40) for FI R2 = 0.7 (RMSEP = 34.1 mg ferulic acid g − 1, RPD = 1.77) for TP R2 = 0.74 (60.0 mmol Trolog kg − 1, RPD = 1.91) for AA | Caporaso et al. (2018) |
Hyperspectral imaging | FI | SVM | 63.3%-90% accuracy | Sanchez et al. (2020) |
Hyperspectral imaging | Classification | PCA, SVM, LDN, KNN | 81.28%-89.10% accuracy | Saeidan et al. (2021) |
Hyperspectral chemical imaging | Total fat content | PLSR | R2 = 0.84, external prediction error of 2.4% for single shelled; \({R}^{2}\)= 0.52, predection error 4.0% for in-shell beans | Caporaso et al. (2021) |
Laser-induced backscattering imaging | Firmness and maturity | PLSR | R2 = 0.755 for 658 nm & R2 = 0.800 for 705 nm for chroma; 90%-95% classification accuracy | Lockman et al. (2019) |
FT-NIR spectroscopy | Fermentation index, pH, and total polyphenol content | ANOVA, LSD | R2 ≥ 0.80 (FI and polyphenol) R2 < 0.80 (pH) | Sunoj et al. (2016) |
FT-NIR spectroscopy | Total fat content | Multivariate Regression, Si-PLS, SVMR, Si-SVMR | RMSEP = 0.015 and Rpre = 0.9708 | Teye and Huang (2015) |
FT-NIR spectroscopy | Classification | LDA, SVM | 100% classification accuracy | Teye et al. (2016) |
FT-NIR spectroscopy | Total fungi count | PLS, Si-PLS,Si-GAPLS, ACO-PLS, CARS-PLS | 0.951 ≤ Rpre ≤ 0.975 & 3.15 ≤ RPD ≤ 4.32 | Kutsanedzie et al. (2018) |
NIR Spectroscopy | Classification, Internal quality | ANOVA, PLSR | RMSE up to 0.99, \({{R}_{c}}^{2}\) up to 0.99 RMSE up to 1.19 \({{R}_{p}}^{2}\) up to 0.97 for whole and ground | Barbin et al. (2018) |
NIR Spectroscopy | Internal quality | PCR, PLSR, MSC, DT, SNV, OSC | R2 = 0.86 RPD = 3.16 (Fat Content), \({R}^{2}\)=0.92 RPD = 3.43 (Moisture Content) | Hayati et al. (2020) |
NIR Spectroscopy | Internal quality | PLSR | R2 = 0.67–0.89, relative error < 10.2% | Hashimoto et al. (2018) |
NIR Spectroscopy | Internal quality | MPLSR, SNV, DT, PLS | R2 = 0.77 (theobromine, R2 = 0.74 (total sugar), R2 = 0.66 (total phenols), R2 = 0.88 (derivatives of epicatechin), R2 = 0.7 (fat), \({R}^{2}\)=0.64 (protein, R2 = 0.82 (husk content) | Hernández-Hernández et al. (2022) |
NIR Spectroscopy | Fermentation Level | PCA | R2 = 0.975, Rpre = 0.935 | Hue et al. (2014) |
NIR Spectroscopy | Classification | CAFS, NBC, ECFS, MCFS | Average accuracy: 99.63% (NBC), 94.92% (ECFS), & 99.63% (MCFS) | Castro et al. (2022) |
Artificial Neural Network | Fermentation index | Coefficient of determination, Bland–Altman plor, Passing-Bablok Regression analysis | No accuracy specified | Leon-Roque et al. (2016) |
Machine Vision and Multiclass SVM Classifier | Classification based on fermentation degree | SVM | 100% accuracy | Yro et al. (2018) |
Digital Imaging | Classification based on morphology | MELS-SVM | 99.705% accuracy | Lawi and Adhitya (2018) |
Multiparametric fluorescence sensor | Anthocyanin, flavanol, chlorophyll and nitrogen balance | PCA, CA | Accuracy was not specified | Tee et al. (2018) |
Electronic nose Systems | Fermentation degree | bootstrap forest, decision tree, boosted tree, ANN, naïve Bayes, k-nearest neighbors | Misclassification rate: 9.4% (bootstrap forest), 12.8% (ANN), 13.6% (boosted tree) | Tan et al. (2019) |
Electronic Nose | Grading | ANOVA | RMSE ~ 1% (KDM), RMSE = 2.9%-9.3% (under different conditions) | Tan and Kerr (2019) |
Electronic Nose | Grading | LDA, SVM, ANN | Overall accuracy: 99% (training dataset), 95% (external-validation dataset) | Hidayat et al. (2019) |
Electronic Nose | Fermentation index | PCA | 97.8% of the total variance | Flórez-Martinez et al. 2020 |
Electronic Nose | Classification | PCA | 95% accuracy | Olunloyo et al. (2012) |
NIR Spectroscopy & Electronic Tongue | Cocoa bean variety Classifications | Multivariate Analysis, PCA | 83%-93% accuracy (single sensor) 100% accuracy (data fusion) | Teye et al. (2014a) |