Ramiah, K. Toole, E. Plant Physiol. Download references. Reprints and Permissions. Physiologia Plantarum Canadian Journal of Genetics and Cytology By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.
Advanced search. Skip to main content. Subscribe Search My Account Login. Abstract THE loss of viability of seeds within one year after harvesting has been reported by Grist 1 from varieties of rice grown in British Guiana and by Ramiah 2 in India, but no explanation has been offered. Rent or Buy article Get time limited or full article access on ReadCube.
References 1 Grist, D. Google Scholar 2 Ramiah, K. Google Scholar 3 Toole, E. After four days, the seeds were removed and were subject to hyperspectral imaging and germination assessment until they naturally dried to their original moisture content. An assembled hyperspectral imaging system covering the range of — nm was applied to acquire images of wheat seeds Figure 1. The data acquisition software could set the speed of the motor, exposure time, and wavelength range.
Each wheat seed, both ventral groove and reverse sides, was placed on the translation stage and transmitted to the camera to be scanned line by line at 1.
- Evidence-Based Geriatric Medicine: A Practical Clinical Guide.
- How To Test The Viability Of Seeds With An Easy Seed Germination Test;
- Shadow Banking?
These images included the linear array scanning by the detector along the y -direction and the movement of the sample in the x -direction. The image acquisition was conducted with Spectral Image software. Then, the calibrated image I was calculated according to the following formula [ 36 ]:. After HSI spectra collection of samples of both sides of the wheat seeds, a germination test was implemented to check for seed viability [ 37 ].
A Petri dish 9 cm was used to secure the seed and the germination paper. After the images were corrected, the background of every image was removed, according to the contrast of the relative reflectance intensity. Then, a series of steps was carried out to extract the spectral data. First, the beginning and ending ranges were omitted from every hyperspectral data file, as they were greatly affected by stochastic noise. Therefore, spectral bands from nm to nm were chosen for further analysis.
Second, wheat seeds were segmented by a threshold imaging procedure at nm to create a mask of the region of interest ROI. The spectra of each pixel from every wheat seed was extracted and averaged. The image segmentation and spectral data extraction were conducted by using ENVI 5. Spectral noise e.
- Seed Viability Chart | High Mowing Organic Non-GMO Seeds.
- Viability and Germination Inhibitor of the Seed of Rice | Nature.
- Viability and Germination Inhibitor of the Seed of Rice.
- How to Test Seed Viability - BBC Gardeners' World Magazine.
- Big Machine: A Novel;
This interference information in the spectral curve decreases the signal: noise ratio and reduces the usefulness of the spectrum data. Extraction of ROIs can improve the data, and pre-processing methods can further maximize usefulness [ 38 , 39 ]. Various pre-processing methods are currently available, such as multiplicative scatter correction MSC [ 40 , 41 ], standard normal variate SNV [ 42 , 43 ], Savitzky-Golay SG , and its derivate [ 16 ].
Savitzky-Golay is used to expose valuable information latent in a spectrum, with the Savitzky-Golay derivative as a frequently used method. It is important to choose an appropriate method for future analysis. In this study, before selecting optimal wavelength, the spectral data were separately preprocessed by the standard normal variate SNV , multiplicative scatter correction MSC , and derivatives that are based on the Savitzky-Golay algorithm with a gap of seven points Figure 2.
The high-dimensional hyperspectral data, comprised of many congruent wavelengths, suffered from multicollinearity and redundancy. Optimal wavelength selection can select the effective wavelengths, which greatly accelerates the computation speed, leading to a time-saving calibration process, and can improve the modelling accuracy. Therefore, optimal wavelength selection is widely applied. The successive projections algorithm SPA is a forward selection algorithm that is regarded as an effective waveband selection method [ 44 ].
Previous research showed that it could minimize the collinearity among variables effectively [ 32 , 45 ].
SPA contains three main steps. Initially, candidate variables are selected according to the maximum projection value on the columns of the spectral matrix [ 46 ]. Then, all of the selected variables are evaluated by root mean square error RMSE. Finally, the variables, which are irrelevant to the properties being predicted, are deleted [ 45 ]. Choosing a reliable classifier is a significant step in building a classification model.
SVM is especially suitable for small datasets with high-dimensional feature spaces. In addition, the grid-search and five-fold cross-validation were used to optimize the parameters. Each spectral dataset was divided into two groups, wheat seeds as calibration sets 75 germinating and 31 non-germinating and 54 seeds as prediction sets 38 germinating and 16 non-germinating.
For the mixture dataset, the data from both sides of seeds were used as calibration sets, and different proportions of data from the ventral groove and reserve sides of seeds were used for the prediction set in the models Table 1. Different proportions of the ventral groove side and the reverse side for constructing the calibration set and prediction set. Models were judged by four indices, which were overall accuracy accuracy , viability accuracy recall , final germination percentage precision , and F-measure. In addition, the initial germination percentage of the prediction set was The overall accuracy classification accuracy of all seeds , viability accuracy classification accuracy of viable seeds , final germination percentage, and F-measure were calculated according to the following formula:.
Hyperspectral information from both sides of wheat seeds was acquired Figure 3. Obvious spectral differences were observed between germinating and non-germinating seeds in the average spectra of the two sides. For the ventral groove side, the values of average spectra of germinating seeds were higher at the wavelengths of — nm, but lower at — nm, when compared with those of non-germinating seeds.
The average spectra of reverse side showed a similar trend. Germinating seeds showed higher average spectra at — nm, while non-germinating seeds exhibited higher average spectra at — nm. These differences indicate that hyperspectral features are influenced by seed vitality and that useful information exists in the hyperspectral imaging of both sides for seed classification.
Average spectra of germination and non-germination seeds for a the ventral groove side and b the reverse side. After preprocessing, the optimal wavelengths were chosen by SPA, which is a novel method to minimize the collinearity among wavebands and select the optimal wavebands. Nearly all of the selected wavelengths were located near particular regions of the spectral range, which indicates some connection between seed reflectance and chemical information. For instance, many selected bands were located near the absorbance bands of plant pigments peaks: carotenoids , nm , chlorophyll a , , and nm , and chlorophyll b , nm [ 50 ].
In addition, bands near nm may correspond to the O—H stretching 2nd overtone and C—H stretching 3rd overtone [ 33 , 50 ]. Bands that exceed nm were disproportionately represented in the selected wavelengths of the reverse side. This could be due to the reverse side containing the embryo with more oil, moisture, and other compounds. Accordingly, the reflectance in narrow wavebands associated with chemical information may be the most useful to classify seed germination.
After using the SPA method to select the spectral wavelength of different spectral datasets, the full wavelength data, as well as the selected wavelength data, were analyzed by PLS-DA and SVM to establish classification models. The results for the ventral groove side spectral dataset are shown in Table 3 and Supplemental Table S1. In these cases, the SVM classification model based on optimum wavelength data with SG method exhibited a highest classification capability with an overall accuracy of This model could increase the germination percentage from Models based on the hyperspectral data of the ventral groove and reverse sides indicated that there were useful spectral data in both sides for identifying viable wheat seeds.
Therefore, we further investigated the discrimination ability of mean and mixture spectra from both sides. This model showed high discrimination ability with an overall classification accuracy of This model could promote the germination percentage of the seed lot from Our results indicated that the model based on the mean spectral dataset showed a lower classification accuracy of overall seeds and viable seeds than that based on the reverse spectral dataset. However, the models using mean spectrum data exhibited a higher ability for improving the final germination percentage of the seed lot than did the optimal model using the reverse side spectrum data.
For modeling that is based on the mixture spectral dataset, the calibration and prediction sets were comprised of equal proportions of the ventral groove side and reverse side data, as described in Table 1.
Seed Viability and Vigour
After screening by this model, the average final germination percentage of the seed lot reached Interestingly, the models performed better with an increasing reverse side ratio, while models performed less accurately with an increase in the number of ventral groove sides Table 4. Results of the models based on different proportions of the ventral groove side and reverse side.
To select the best pre-processing methods for reducing noise interference in the spectral data and to enhance model accuracy, different preprocessing methods were applied to investigate the model performance. The results showed that the SG, SNV, and MSC methods had different effects on the performance of the models that are based on different spectral datasets.
How to Test Seeds for Viability
Thus, models must be constructed to employ specific preprocessing methods for different datasets [ 25 ]. In addition, our results indicated that the SPA algorithm was suitable for selecting the important variables and for modeling the viability screening of wheat seeds. Constructing the model using the particular wavelengths selected by SPA reduced hyperspectral imaging and processing time [ 33 , 51 ].
Most optimum models that were based on different spectral datasets were produced by using pre-processing methods and particular wavelengths selected by SPA. Our work indicated that these two classifiers could distinguish the viability of wheat seeds.
- How To Test The Viability Of Seeds With An Easy Seed Germination Test.
- How to Test Seeds for Viability.
- How Long Will Old Seeds Stay Viable??
Overall classification accuracy is a weighted average of viability and non-viability accuracy, which is influenced by both the accuracy of determining the viability and non-viability, and by the frequency of viable and non-viable seeds. In this work, the original germination percentage of the seed lot was Therefore, high overall accuracy cannot completely reflect the performance of the models.
When selecting a practical model for seed viability classification, not only the overall accuracy, but also the accuracy of viable seeds and the enhancing effect of final germination percentage should be considered as selection criteria. In this model, high overall accuracy and viable seeds were obtained by only using selected wavelengths. This system greatly reduced the number of used wavelengths from to 16, and could be applied to develop multispectral screening machines for seed companies and farmers.
In addition to providing high classification capability, it is cost-effective, requiring only one scanner to acquire spectral information. Acquisition of average spectral data requires two cameras to simultaneously screen with a specific machine that is needed to keep the seeds in the appropriate orientation for reverse-side information acquisition, negatively affecting the speed and cost of production.
The methodology of hyperspectral image acquisition, pre-processing, and of wavelength and classifier selection was established for the classification of viable and non-viable wheat seeds. Likewise, the SPA method, which extracted the most effective wavebands and optimized models, was suitable for wheat seed characterization according to viability. Furthermore, after screening by this model, the average final germination percentage of the seed lot could exceed Table S1. Table S2. Table S3. National Center for Biotechnology Information , U. Journal List Sensors Basel v.
Sensors Basel. Published online Mar 8. Find articles by Tingting Zhang.
Viability of Seeds
Find articles by Bin Zhao. Find articles by Ranran Wang. Find articles by Mingliu Li. Find articles by Jianhua Wang. Find articles by Qun Sun. Author information Article notes Copyright and License information Disclaimer. Received Jan 15; Accepted Mar 6. This article has been cited by other articles in PMC.
Associated Data Supplementary Materials sensorss Introduction Seeds are the basis of the agricultural industry [ 1 ]. Materials and Methods 2. Seed Preparation Dry seeds of the wheat cultivar Dunmaiwang were purchased from a local market Jinan, China in October Hyperspectral Imaging System An assembled hyperspectral imaging system covering the range of — nm was applied to acquire images of wheat seeds Figure 1. Open in a separate window.
Figure 1. Image Acquisition and Calibration Each wheat seed, both ventral groove and reverse sides, was placed on the translation stage and transmitted to the camera to be scanned line by line at 1. Germination Assessment After HSI spectra collection of samples of both sides of the wheat seeds, a germination test was implemented to check for seed viability [ 37 ]. Spectral Data Extraction After the images were corrected, the background of every image was removed, according to the contrast of the relative reflectance intensity.
Spectra Preprocessing Spectral noise e. Figure 2. Optimal Wavelength Selection The high-dimensional hyperspectral data, comprised of many congruent wavelengths, suffered from multicollinearity and redundancy. Development of Classification Models Choosing a reliable classifier is a significant step in building a classification model. Table 1 Different proportions of the ventral groove side and the reverse side for constructing the calibration set and prediction set.
Results 3. Spectral Characteristics Hyperspectral information from both sides of wheat seeds was acquired Figure 3. Figure 3. Optimal Wavelengths Selected by the SPA Algorithm After preprocessing, the optimal wavelengths were chosen by SPA, which is a novel method to minimize the collinearity among wavebands and select the optimal wavebands.
Table 2 Selected wavelengths by successive projections algorithm SPA. Table 3 The best results of models based on each spectral dataset.
Datasets Pre-Processing No. Table 4 Results of the models based on different proportions of the ventral groove side and reverse side. Discussion To select the best pre-processing methods for reducing noise interference in the spectral data and to enhance model accuracy, different preprocessing methods were applied to investigate the model performance. Conclusions The methodology of hyperspectral image acquisition, pre-processing, and of wavelength and classifier selection was established for the classification of viable and non-viable wheat seeds.
Click here for additional data file. Conflicts of Interest The authors declare no conflict of interest. References 1. Huang M. Classification of maize seeds of different years based on hyperspectral imaging and model updating. Dumont J. Thermal and hyperspectral imaging for Norway spruce Picea abies seeds screening. Soybean seed viability and changes of fatty acids content as affected by seed aging. Dong K.