Skip to main content

Advertisement

Table 1 Summary of selected references applying hyperspectral imaging to seed classification and seed grading

From: Hyperspectral imaging for seed quality and safety inspection: a review

Seed Spectral rangea Varieties Sample numbers Features Signal mode Data analysis strategies Main application type Classification result (highest accuracy) References
Spectra/image Extraction/selection methods Analysis level Classification/regression methods
Barley, wheat and sorghum 1000–2498 1 variety of each kind of grain 150 of each kind of grain Spectra PCA Reflectance PWb prediction map and OWc (single kernels) Grain topography classification Manley et al. [19]
Black bean 390–1050 (501–1000) 3 300 Spectra and image SPA, PCA, GLCM Reflectance OW (single kernels) PLS-DA, SVM Variety classification 98.33% (PLS-DA) Sun et al. [15]
Grape seed 897–1752 (914–1715) 3 varieties, two growth soil 56 Spectra PCA Reflectance OW (single kernels), PW PCA and prediction map GDA Assess Stage of maturation of grape seeds > 95% Rodríguez-Pulido et al. [22]
Grape seed 874–1734 (975–1646) 3 43,357 Spectra and image PCA Reflectance OW (single kernels) SVM Variety classification 94.30% Zhao et al. [26]
Maize 874–1734 (972–1642) 2 (transgenic and non-transgenic) 2100 Spectra PCA, CARS Reflectance PW PCA and prediction map, OW (single kernels) PLS-DA, SVM Transgenic and non-transgenic classification 99.5% (PLS-DA) Feng et al. [24]
Maize 400–1000 4 varieties, 3 crop years 3600 Spectra no Reflectance OW (single kernels) LS-SVM Variety classification 91.50% Guo et al. [12]
Maize 400–1000 4 varieties, 3 crop years 2000 Spectra no Reflectance OW (single kernels) LS-SVM Variety classification 94.80% He et al. [13]
Maize 400–1000 4 varieties, 3 crop years 2000 Spectra no Reflectance OW (single kernels) LS-SVM Variety classification 94.40% Huang et al. [11]
Maize 400–1000 (400–1000) 17 1632 Spectra and image PCA, SPA, GLCM, MDS Reflectance OW (single kernels) LS-SVM Variety classification 94.40% Huang et al. [17]
Maize 1000–2500 18 36 Spectra and image PCA Reflectance OW (single kernels), PW PCA and prediction map PLS-DA Textural, vitreous, floury and the third type endosperm 85% (PLS-DA) Manley et al. [20]
Maize 975–2570 (1101–2503) 3 hardness 115 Spectra and image PCA Reflectance PW PCA and prediction map, OW (single kernels) PLS-DA Hardness classification 97% (PLS-DA) Williams and Kucheryavskiy [18]
Maize 874–1734 (924–1657) 14 1120 Spectra joint skewness-based wavelength selection Reflectance OW (single kernels) LS-SVM Variety classification 98.18% Yang et al. [7]
Maize 874–1734 (975–1646) 3 12,900 Spectra and image PCA Reflectance OW (single kernels) SVM, RBFNN Variety classification 93.85% (RBFNN) Zhao et al. [25]
Maize 380–1030 (500–900) 6 330 Spectra and image PCA, KPCA, GLCM Reflectance OW (bulk samples) LS-SVM, BPNN, PCA, KPCs Classes classification 98.89% (PCA-GLCM-LS-SVM) Zhang et al. [78]
Rice 390–1050 (500–951) 4 origins 240 Spectra and image PCA, GLCM Reflectance OW (single kernels) SVM Variety classification 91.67% Sun et al. [16]
Rice 874–1734 (1039–1612) 4 225 Spectra PLS-DA, PCA Reflectance PW PCA and OW (bulk samples) KNN, PLS-DA, SIMCA, SVM, RF Seed cultivars classification 100% (SIMCA, SVM, and RF) Kong et al. [5]
Soybean, maize and rice 400–1000 (400–1000) 3 of each kind of seed 225 of each kind of seed Spectra neighborhood mutual information Reflectance OW (single kernels) ELM, RF Variety classification 100% (ELM) Liu et al. [8]
Waxy corn 400–1000 (430–980) 4 600 Spectra and image SPA, GLCM Reflectance OW (single kernels) PLS-DA, SVM Variety classification 98.2% (SVM) Yang et al. [14]
Wheat 960–1700 (960–1700) 8 2400 Image WT, STEPDISC, PCA Reflectance PW and OW (bulk samples) BPNN, LDA, QDA Classes classification 99.1% (LDA) Choudhary et al. [79]
Wheat 960–1700 (960–1700) 8 2400 Spectra STEPDISC Reflectance OW (bulk samples) LDA, QDA, Standard BPNN, Wardnet BPNN Variety classification 94–100% (LDA) Mahesh et al. [6]
Wheat 960–1700 (960–1700) 5 2500 Spectra STEPDISC Reflectance PW PCA and OW (bulk samples) LDA, QDA Classes classification 90–100% (LDA) Mahesh et al. [9]
  1. aThe spectral range without brackets relates to the range acquisition of instrument, while the spectral range in brackets represents the spectral range for practical analysis
  2. bPW means pixel-wise analysis, which is the analysis on the pixels
  3. cOW means objective-wise analysis, which means the analysis on ROIs (ROI can be bulk, single kernel or self-defined)