-
The results of the standard germination test showed that the seed germination potential and germination percentage increased significantly with the increase of maturity. The germination potential increased from 7% to 91%, and the germination percentage of seeds at milk-ripening stage was only 46%, at wax-ripening stage 56% and at full-ripening stage it could rise to 94% (Fig. 1a). On the other hand, the seed germination potential and germination percentage decreased significantly with increasing length of post-harvest storage. Seeds harvested in 2019 had a germination percentage of 97% and a germination potential of 89%. The germination percentage of seeds harvested in 2017 had dropped to 8% after storage for four years, with only 1% germination potential, while the current germination potential and germination percentage of seeds harvested in 2014 were 0% (Fig. 1b).
Figure 1.
Germination situation of smooth bromegrass seeds under different conditions. Data in (a) and (b) were the germination potential and germination percentage of smooth bromegrass seeds at different maturity levels and harvest years under standard germination test, respectively. Data in (c) and (d) were the germination percentage of smooth bromegrass seeds at different maturity levels and harvest years after artificial accelerated aging, respectively. Different lowercase letters represent significant difference between different treatments during the experimental period (p < 0.05). Bars represent the mean values of four replicates ± standard deviation (SD).
The results of artificial accelerated aging test of seed vigour showed that there were significant differences in germination percentage among three different maturity levels. The seeds at the full-ripening stage had the highest germination percentage and the seeds at the wax-ripening stage had the lowest germination percentage (Fig. 1c). For the seeds of three different harvest years, the germination percentage of seeds harvested in 2014 and 2017 were 0%, which were low vigour seeds, and seeds harvested in 2019 were high vigour seeds (Fig. 1d).
Chlorophyll fluorescence measurement
-
Fluorescence intensity maps of chlorophyll a and b captured by multispectral showed that among the seeds of different maturity levels, seeds at the milk-ripening and wax-ripening stages showed higher value of chlorophyll a, significantly higher than the full-ripening stage, and seeds at the full-ripening stage had the highest value of chlorophyll b (Fig. 2a & c). While among the seeds from different harvest years, the seeds harvested in 2019 had the highest content of chlorophyll a and chlorophyll b, and the values of chlorophyll a and chlorophyll b showed a gradient decrease with increasing storage year (Fig. 2b & d).
Figure 2.
Chlorophyll a and chlorophyll b fluorescence intensity of smooth bromegrass seeds at different maturity levels and harvest years. Data in (a) and (b) were chlorophyll a fluorescence intensity of seeds at different maturity levels and harvest years, respectively. Data in (c) and (d) were chlorophyll b fluorescence intensity of seeds at different maturity levels and harvest years, respectively. Different lowercase letters represent significant difference between different treatments during the experimental period (p < 0.05). Bars represent the mean values of four replicates ± SD.
Analysis of morphological and spectral data
-
Fourteen morphological characteristics of each seed lots were extracted by multispectral imaging, and seeds of different maturity levels and different harvest years were compared separately. The results showed significant differences in 11 indicators among seed lots of different maturity levels (p < 0.05), with significant differences in area, length-width ratio, shape parameters, color parameters and saturation among seed lots at different maturity levels, and in hue indicators, the seeds at milk-ripening stage were significantly higher than the seeds at wax-ripening stage and the full-ripening stage. There were 10 significant differences among seed lots in different harvest years (p < 0.05), among which there were significant differences in CIELab L* and CIELab B* of color parameters, saturation and hue, as well as in length-width ratio and shape parameters (Table 1).
Table 1. Comparative analysis of morphological characteristics of smooth bromegrass seeds with different maturity levels and different harvest years.
Characteristic Different maturity Different harvest years Milk-ripening stage Wax-ripening stage Full-ripening stage 2019 2017 2014 Area (mm2) 12.61 ± 2.095c 12.89 ± 1.777b 13.45 ± 1.668a 15.87 ± 2.675b 17.74 ± 2.845a 16.14 ± 2.512b Length (mm) 9.87 ± 0.901a 9.65 ± 0.904b 9.89 ± 0.744a 12.55 ± 1.793b 13.44 ± 2.004a 12.58 ± 1.843b Width (mm) 1.81 ± 0.278c 1.90 ± 0.211b 2.02 ± 0.212a 2.19 ± 0.282b 2.31 ± 0.283a 2.16 ± 0.253b Length-width ratio 0.18 ± 0.029c 0.20 ± 0.024b 0.20 ± 0.022a 0.18 ± 0.029 0.17 ± 0.027 0.18 ± 0.039 Compactness Circle 0.18 ± 0.028c 0.19 ± 0.023b 0.20 ± 0.019a 0.17 ± 0.025a 0.17 ± 0.021b 0.18 ± 0.036a Compactness Ellipse 0.99 ± 0.008b 0.99 ± 0.006a 0.99 ± 0.008b 0.96 ± 0.029b 0.96 ± 0.025b 0.97 ± 0.022a BetaShape_a 1.73 ± 0.221c 1.79 ± 0.182b 1.92 ± 0.237a 3.30 ± 1.878 3.33 ± 2.059 3.07 ± 1.415 BetaShape_b 1.58 ± 0.199c 1.69 ± 0.174b 1.78 ± 0.208a 2.74 ± 1.169 2.75 ± 1.032 2.61 ± 0.847 Vertical Skewness −0.07 ± 0.051c −0.04 ± 0.034a −0.06 ± 0.046b −0.10 ± 0.088 −0.09 ± 0.076 −0.09 ± 0.07 CIELab L* 46.28 ± 2.959b 44.72 ± 3.395c 47.06 ± 2.306a 51.43 ± 2.578a 51.06 ± 2.442b 48.42 ± 2.047c CIELab A* 6.17 ± 1.402b 4.96 ± 1.443c 7.94 ± 0.787a 6.26 ± 0.73b 7.71 ± 0.847a 7.67 ± 0.823a CIELab B* 17.78 ± 1.682b 12.58 ± 3.04c 18.58 ± 1.32a 17.57 ± 1.658c 20.18 ± 1.56a 19.68 ± 1.879b Saturation 19.06 ± 1.844b 13.50 ± 3.132c 20.66 ± 1.392a 18.80 ± 1.738c 21.79 ± 1.674a 21.33 ± 1.932b Hue 1.23 ± 0.066a 1.17 ± 0.112b 1.17 ± 0.03b 1.23 ± 0.033a 1.21 ± 0.028b 1.19 ± 0.109c Note: Different lowercase letters in the same line indicate significant differences, while the same letters indicate no significant differences (p < 0.05). The number of repetitions for this experiment was n = 4 for each seed lot samples. Through multispectral data collection and analysis, the average reflectance of smooth bromegrass seeds at 570−645 nm in wax-ripening stage was completely separate from that of milk-ripening stage and full-ripening stage, and the average reflectance of smooth bromegrass seeds at 780−970 nm in full-ripening stage was differed considerably from that of milk-ripening stage and wax-ripening stage (Fig. 3a). The average reflectance of smooth bromegrass seeds harvested in 2014 at 450−690 nm was completely separate from those smooth bromegrass seeds harvested in 2017 and 2019, and the mean reflectance of smooth bromegrass seeds harvested in 2019 at 940−970 nm differed significantly from those in 2014 and 2017 (Fig. 3b).
Figure 3.
Mean spectral reflectance of smooth bromegrass seeds at (a) different maturity levels and (b) harvest years.
Multivariate analysis
-
The results of the principal component analysis based on morphological and spectral features showed that for seeds at different maturity levels, the first two principal components explained only 53.6% of the original variance, and there was more overlap in the 95% confidence ellipses in the PC1/PC2 two-dimensional plots for seed samples at milk-ripening and wax-ripening stages. Comparatively speaking, there was less overlap between the seeds at full-ripening stage and seeds at the other two stages (Fig. 4a). For seeds from different harvest years, the first two principal components explained only 54.5% of the original variance and had a large overlap, this indicated that there was a high degree of similarity among samples and that PCA was not effective in clustering smooth bromegrass seeds of different maturity levels and different years (Fig. 4b). In addition, all spectral data were positively correlated with the principal component 1(PC1), with 490 nm, 515 nm, 540 nm and 690 nm contributing more to PC1. Morphological character data saturation, color parameter CIELab A* and CIELab B* were positively correlated with principal component 2(PC2) of smooth bromegrass seeds at different maturity levels (Supplemental Fig. S4a), while negatively correlated with PC2 of smooth bromegrass seeds at different harvest years (Supplemental Fig. S4b). In terms of the separation effect of smooth bromegrass seeds with different maturity levels, LDA could clearly distinguish the smooth bromegrass seeds of full-ripening, milk-ripening and wax-ripening stage, with the best distinction of full ripening stage seeds (Fig. 4c). For smooth bromegrass seeds of different harvest years, although the seeds harvested in 2014 and 2017 both had low vigour, they also had a good differentiation effect, and the seeds harvested in 2019 had the best discrimination with other seeds (Fig. 4d).
Figure 4.
Multivariate analysis based on multispectral data of smooth bromegrass seeds. Principal component analysis of smooth bromegrass seeds at (a) different maturity levels and (b) harvest years. LDA model diagram of smooth bromegrass seeds at (c) different maturity levels and (d) harvest years.
The LDA model had the best prediction effect on seed differentiation at different maturity levels and harvest years compared to the RF and SVM. The accuracy of LDA could reach 94.2% (milk-ripening vs wax-ripening stages), 99.2% (wax-ripening vs full-ripening stages) and 99.6% (milk-ripening vs full-ripening stages), respectively (SVM was 93.8%, 97.9% and 99.2%, RF was 87.5%, 98.3% and 97.1%), with an average accuracy of 97.7%, and the sensitivity, specificity and accuracy could also reach more than 96.0% on average (Table 2). At same time, LDA could also distinguish seeds at different harvest years, and the accuracy was able to reach 90.0% (2014 vs 2017), 99.2% (2017 vs 2019) and 97.5% (2014 vs 2019), respectively (SVM was 86.3%, 96.3% and 95.0%, RF was 77.9%, 89.2% and 85.8%), with an average accuracy of 95.6%, and the sensitivity, specificity and precision were high, which could reach more than 95.0% on average (Table 2).
Table 2. Prediction of smooth bromegrass seeds with different maturity levels and harvested in different harvest years by LDA, SVM and RF models.
Model Index Different maturity Different harvest year M vs W W vs F M vs F 2014 vs 2017 2017 vs 2019 2014 vs 2019 LDA Sensitivity (%) 96.6 99.2 99.2 91.6 98.3 96.6 Specificity (%) 91.7 99.2 100.0 88.4 100.0 98.3 Precision (%) 92.0 99.2 100.0 88.6 100.0 98.3 Accuracy (%) 94.2 99.2 99.6 90.0 99.2 97.5 SVM Sensitivity (%) 95.8 98.3 100.0 89.9 96.6 95.0 Specificity (%) 91.7 97.5 98.3 82.6 95.9 95.0 Precision (%) 91.9 97.5 98.3 83.6 95.8 95.0 Accuracy (%) 93.8 97.9 99.2 86.3 96.3 95.0 RF Sensitivity (%) 85.7 98.3 97.5 81.5 89.9 89.1 Specificity (%) 89.3 98.3 96.7 74.4 88.4 82.6 Precision (%) 88.7 98.3 96.7 75.8 88.4 83.5 Accuracy (%) 87.5 98.3 97.1 77.9 89.2 85.8 Note: M stands for milk-ripening stage, W stands for wax-ripening stage and F stands for full-ripening stage. Prediction of seed germination
-
In addition, we also predicted the germination of each smooth bromegrass seed according to the nCDA model in VideometerLab4 software, and the accuracy of the model prediction was verified and evaluated by the actual results after the standard germination test was completed. The standard germination results of seed samples at different maturity levels and harvest years were classified and recorded separately according to dead seeds (D), fresh ungerminated seeds (F), abnormal seedlings (A) and normal seedlings (N), and the respective prediction accuracies were calculated separately.
The calculated nCDA prediction results showed that, regarding seed samples of different maturity levels, the prediction accuracy of dead seeds with no vigour could reach 93.7%, normal seedlings with high vigour could reach 91.1%, abnormal seedlings and fresh ungerminated seeds could reach 97.1% and 97.5%, but the sum of abnormal and normal seedlings, i.e., seeds capable of germination (A + N), only reached 85.8%. For seed samples from different harvest years, the prediction accuracy of dead seeds was 91.3%, the prediction accuracy of normal seedlings could reach 91.3%, the prediction accuracy of abnormal seedlings and fresh ungerminated seeds could reach 92.6% and 98.3%, but the prediction accuracy of seeds that could germinate (the sum of abnormal seedlings and normal seedlings) could reach 94.0% (Table 3).
Table 3. Germination prediction of smooth bromegrass seeds with different maturity levels and harvest years based on nCDA.
Sample Classification Actual number of CCS Correctly predicted number of CCS Actual number of RSS Correctly predicted number of RSS Accuracy of prediction (%) Maturity level D 301 247 899 877 93.7% F 7 5 1193 1165 97.5% A 109 70 1091 1095 97.1% N 783 769 417 324 91.1% A + N 892 811 308 218 85.8% Harvest year D 766 691 434 404 91.3% F 3 3 1197 1176 98.3% A 12 9 1188 1102 92.6% N 419 390 781 705 91.3% A + N 431 386 769 742 94.0% Note: CCS stands for Corresponding classification samples, RSS stands for Remaining sorted samples. N stands for normal seedlings, A stands for abnormal seedlings, F stands for fresh ungerminated seeds and D stands for dead seeds. Overall, the SVM model was better than both LDA and RF models in predicting the germination of smooth bromegrass seeds harvested in different years. However, the accuracy of both LDA and SVM models for predicting normal seedlings, abnormal seedlings, fresh ungerminated seeds and dead seeds was above 90.0%, and the accuracy of predicting abnormal seedlings and fresh ungerminated seeds was the same, the prediction accuracy of LDA was up to 98.3% and SVM was up to 99.2%. For the seed samples of different maturity levels, it was similar to the prediction of germination of seeds harvested in different years. The accuracy of LDA model and SVM model for predicting fresh ungerminated seeds was higher, which were 98.9% and 99.2%, respectively. And the prediction accuracy of LDA model and SVM model for abnormal seedlings was more than 90.0%. However, for normal seedlings and dead seeds, the prediction accuracy of both models was lower. The prediction accuracy of normal seedlings was less than 70.0%, and that of dead seeds was less than 80.0% (Table 4).
Table 4. Germination prediction of smooth bromegrass seeds with different maturity levels and harvested in different years by LDA, SVM and RF models.
Model Index Different maturity Different harvest years N A F D N A F D LDA Sensitivity (%) 79.9 0 0 12.4 88.4 0 0 96.5 Specificity (%) 40.5 99.7 99.7 94.5 96.5 99.2 99.2 87.1 Precision (%) 72.6 0 0 42.3 93.4 0 0 92.8 Accuracy (%) 66.7 91.7 98.9 74.2 93.6 98.3 98.3 93.1 SVM Sensitivity (%) 80.8 0 0 0 86.0 0 0 97.4 Specificity (%) 40.5 100.0 100.0 100.0 97.0 100.0 100.0 84.1 Precision (%) 72.8 NA NA NA 94.1 NA NA 91.4 Accuracy (%) 67.2 91.9 99.2 75.3 93.1 99.2 99.2 92.5 RF Sensitivity (%) 79.1 0 0 12.4 69.0 0 0 93.0 Specificity (%) 39.7 99.7 100.0 95.6 93.1 100.0 100.0 72.0 Precision (%) 72.1 0 NA 47.8 84.8 NA NA 85.1 Accuracy (%) 65.8 91.7 99.2 75.0 84.4 99.2 99.2 85.3 Note: N stands for normal seedlings, A stands for abnormal seedlings, F stands for fresh ungerminated seeds and D stands for dead seeds. -
Based on multispectral technology, by extracting morphological data and spectral information of seeds and combining with LDA model, the seed vigour of smooth bromegrass could be tested more accurately, and the seed viability and germination percentage of smooth bromegrass could also be tested more accurately by combining with nCDA model. The method above was applicable to the determination of seeds with different vigour levels caused by different maturity levels in the process of seed formation and development, and also to the determination of seeds with different vigour levels caused by natural aging in different storage years after physiological maturity, which has good applicability and representativeness. In conclusion, the results of this study confirmed the feasibility of multispectral imaging techniques combined with multivariate analysis methods to rapidly achieve differentiation of smooth bromegrass seed vigour, which has a good prospect in non-destructive testing for seed vigour.
-
About this article
Cite this article
Yang S, Zhang S, Yi K, Wei K, Zeng H, et al. 2023. Rapid non-destructive testing of smooth bromegrass (Bromus inermis) seed vigour using multispectral imaging. Grass Research 3:12 doi: 10.48130/GR-2023-0012
Rapid non-destructive testing of smooth bromegrass (Bromus inermis) seed vigour using multispectral imaging
- Received: 15 February 2023
- Accepted: 19 May 2023
- Published online: 03 July 2023
Abstract: Seed vigour is an important indicator to evaluate the seed quality, and the testing of seed vigour is crucial during the growth and development of seeds. Multispectral imaging is an emerging non-destructive testing technology that has been gradually utilized in the field of seed quality testing in recent years. In this study, we used multispectral imaging to obtain image and spectral information of smooth bromegrass seeds with different maturity levels and different harvest years, and combined with five multivariate analysis methods of principal component analysis (PCA), linear discriminant analysis (LDA), support vector machine (SVM), random forest (RF) and normalized canonical discriminant analysis (nCDA) to distinguish and estimate. Results showed that LDA could predict the seed vigour of smooth bromegrass more accurately, and the accuracy of seed prediction could reach 94.2%−99.6% for different maturity levels and 90.0%−99.2% for different harvest years, which was the best differentiation model. On the prediction of seed germination, the accuracy of the LDA model in predicting normal seedlings of smooth bromegrass seeds in different harvest years (93.6%) was higher than different maturity levels (66.7%). The prediction accuracy of nCDA could reach 93.7% and 91.3% for dead seeds with different maturity levels and different harvest years, and 91.1% and 91.3% for high vigour seeds, respectively. The above results confirmed the feasibility of the multispectral imaging techniques combined with the multivariate analysis methods for rapid and non-destructive differentiation of seed vigour of smooth bromegrass, and it also showed good application prospects in the seed vigour testing field.