How Can the Engineering Parameters of the NIR Grader Affect the Efficiency of Seed Grading?

Abstract

The automated grading of Scots pine seeds in the near-infrared wavelength region (NIR grading) is a starting point for further actions, such as coating and priming. This reduces the time and financial costs and increases the accuracy of seed viability classification compared to invasive techniques. The NIR-based wave reflected from each pine seed must be detected and processed with sufficient accuracy. To focus the reflected beam, we used fiber-optic Bragg grating, a Bragg mirror, and diffraction grating. For each focusing option based on the DOE matrix, one experiment of 20 runs (n = 20) and three replicas (m = 3) in each run was conducted. In each replica, we used 100 conditioned and 100 non-conditioned seeds (N_C + N_NC = 200) selected randomly from five samples weighing 50 g from a seedlot weighing 1 kg extracted from cones collected from a natural tree stand. Three experiments were conducted on the NIR grading of Scots pine seeds using an optoelectronic device. An adequate DOE regression model of the grading efficiency function was obtained. The functions included the following arguments: angle of incidence of the optical beam, NIR wavelength reflected from the seed, and height of the seed pipeline. The influence of the inclination angle of the light source relative to the plane of pine seed movement on the grading quality prevails over other factors. The NIR grading of Scots pine seeds allows the separation of seeds according to the viability index, which is important, since dead petrified seeds (possibly up to 25%) may occur in the seed batch, which cannot be eliminated by either seed size or mass. The peak of NIR grading is achieved by combining the average grader engineering parameters: 968–973 nm for the wavelength and 44–46 degrees for the inclination angle of the reflected beam at a seed pipe size of 0.18–0.23 m.

1. Introduction

The choice of Scots pine (Pinus sylvestris L.) is due to the prevalence of this forestry culture as the main one in single and mixed forest stands along the gradient between the 50th and 60th parallels from North America to Eurasia (native only to Eurasia). The decomposition of a modern set [1] of technological processes of reforestation [2] focuses on grading seeds using optical detection [3,4].

The use of optical beams is important for the assessment [5], preparation [6], and non-destructive quality control [7,8,9,10] of reproductive material [11,12,13,14], including Scots pine [15,16,17,18,19]. The main driving factor for this study is the different absorption ability of a conditioned and substandard single seed of Scots pine of an optical beam in the near-infrared (NIR) region [20]. The Scots pine seed spectrometric feature in the NIR region is clearly detected and correlates with the seeds’ viability [4,21,22]. For all types of direct seeding [23] (aerial [24,25,26], ground based [27] with controlled mycorrhization [28], gravity [29]), the use of seeds graded according to the NIR feature will reduce the sowing costs and rates by increasing the seeds’ germination [30] and the seedling growth intensity [13,31]. The seeds’ quality in this case is determined by the accuracy of the optoelectronic [32] and mechatronic [33,34] systems of the separating apparatus [35] and the correctness of their complexing and alignment. This can contribute to improving the “rational choice of technological operations using mechanized and automated technical devices after making a decision regarding the restoration of the forest landscape (FLR technology)” [1].

Of particular interest in the implementation of the process are three technological operations (orientation, diagnosis, and calibration of seeds) carried out in the device. The orientation operation should ensure a uniform supply of Scots pine seeds, one at a time, to the working space of the separating device. Therefore, when designing a hopper, it is advisable to first investigate the angles of the natural slope of the seed pile, as well as the seeds’ shape and the inter-shift interaction in the hopper [36]. The diagnostic operation should provide fast digital analysis of Scots pine seeds using optoelectronic devices of the developed configuration [4,35,37,38,39,40,41]. During the operation, non-destructive testing of the shape, size, and spectrometric characteristics of the seeds is possible, while simultaneously changing the thermophysical properties. The basis of the operation is a clear detection of seeds according to a given spectrometric criterion [4]. The calibration operation should ensure timely and clear separation of Scots pine seeds into fractions according to the control signal of the diagnostic system, in accordance with the specified analyzing algorithm for managing the quality of separation, the minimum possible mechanical impact on the seeds, and high speed [42] relative to the other two systems.

The primary information about the state of seeds is carried by the reflected beam of the following NIR wavelength λ_r(NIR) focused into the photodetector. The reflected beam is focused on the photodetector by the following elements implemented in the corresponding inventions: fiber-optic Bragg gratings (I type) [35], Bragg mirrors (II type) [39], or a diffraction grating (III type) [41]. Presumably, there should be no obstacles to scaling this study to other types and collection sites of gymnosperm seeds, since the spectral line of many of these seeds in this wavelength range has a characteristic extremum. However, in our opinion, it would be more correct to obtain and refine the spectral characteristic for each type of seed by entering it into the FRM-Library database [14] in order to synchronize it with the software and hardware complex of seed grading.

The aim of the study is to determine the optimal basic engineering parameters (λ_r(NIR)—the NIR wavelength of the reflected beam; α—the inclination angle from the light source; h_sp—the seed pipe height) of the NIR grader and to assess the adequacy of the corresponding regression models for predicting the total grading efficiency function y (λ_r(NIR), α, h_sp).

2. Materials and Methods

2.1. Seed Sampling

A seedlot (m = 1 kg) of Scots pine (P. sylvestris), obtained from cones collected in the fall of 2019 from a natural tree stand (latitude 50.462169, longitude 40.096446, altitude 83 m a.s.l., Voronezh region, Russia), was taken from the breeding and seed-growing center. Seed pretreatment was carried out according to the standard method [12] using BCC AB equipment (Sweden): extraction from cones and seed drying (dry chamber DL-1200), wet de-winging (Dewinger 800), and cleaning and gravity sorting (gravity separator Mini-Series). Five samples (m = 50 g) were randomly selected from the seedlot, in each of which two groups of 100 seeds were selected using radiography: conditioned (C) group and non-conditioned (NC) group. The C group was identified as viable seeds characterized by the presence of an endosperm and an embryonic axis. The NC group was identified as seeds characterized by the presence of an endosperm and the absence of an embryonic axis or the absence of both an endosperm and an embryo. Further, spectrograms in the NIR region were taken from the C and NC samples, and a region from 960 to 980 nm was established in which it was possible to differentiate the C and NC samples by reflected radiation amplitude [30] A according to

Table 1. Detection criteria for Scots pine seeds’ NIR grading using an NIR spectrometer.

Spectrometric Group	NIR Wavelength, nm	Radiation Amplitude A
C	960–980	<60%
NC	960–980	≥60%

.

2.2. Grader’s Parameters

With optoelectronic grading of seeds under real conditions, frequency selection and determination of the amplitude of the optical beam are always carried out with a certain error. Based on the physical meaning of this problem, the use of design of experiment (DOE) methods seems to be most adequate for constructing an analyzing plan matrix. DOE is a procedure for selecting the number and conditions of experiments necessary and sufficient to solve a task with the required accuracy [43]. At the same time, the following are essential: the desire to minimize the total number of experiments; simultaneous variation of all variables that determine the process, according to special rules—algorithms; and the use of a mathematical apparatus that formalizes many actions of the experimenter [43]. To evaluate the effectiveness of grading seeds, a constructive scheme was used (Figure 1) with the detection of an optical beam diffusely reflected from a single seed of Scots pine.

The performance of this design scheme involves the time of operation of the optoelectronic and mechanical elements. The maximum response time [32] refers to a stepper motor, so previously, Novikov et al. (2019) [42] considered the single-seed mechanical movement. That is, the process of detecting a single seed can be carried out during the free movement of the seed in the gravitational field (for a height of 20 cm without taking into account air resistance, the fall time of a single seed at a speed of 1.981 m s⁻¹ is about 0.2 s). At the same time, the productivity of the system in the slowest version will be 1000 seeds in 200 s, or, taking into account the mass of 1000 seeds, about 5 g–90 g h⁻¹. If each subsequent seed is fed after a time sufficient for the operation of a stepper motor (0.01 s), without waiting for the previous seed to reach the corresponding receiver, the productivity can increase by 5 times or more, that is, about 450 g h⁻¹. When used in the mass of an industrial process, the quality of the seed should prevail over the quantity. On average, on an industrial scale, 16000 h are required annually for processing a potentially possible harvest of 8000 kg of Scots pine seeds for one implementation of the system. Thus, having implemented 1000 grading systems with such a constructive scheme, it is possible to spend only 2 working days of 8 h.

2.3. DOE—Optimizing the Grader’s Parameters

To optimize the grader parameters, a full-factor experiment was implemented using a central compositional uniform rotatable plan of the second order [44]. Each factor in the experiment was varied at five levels according to

Table 2. Values of the factors and the levels of their variation in the implementation of the experiment plan for Scots pine seeds’ NIR grading.

Designation		Levels of Factor Variation					Interval
Natural	Code	–α	–1	0	+1	+α
λr(NIR), nm	x1	960	964	970	976	980	6
α, °	x2	35	39	45	51	55	6
hsp, m	x3	0.10	0.14	0.20	0.26	0.30	0.06
$x_1=\frac{{\lambda }_{r\left(\mathrm{NIR}\right)}-970}{6}$; $x_2=\frac{\alpha -45}{6}$; $x_3=\frac{h_{sp}-0.2}{0.06}$

. A priori, the main design factors of the grader affecting the response of the NIR grading process [45] were identified: the NIR wavelength, λ_r(NIR) (x₁); the inclination angle from the light source, α (x₂); and the seed pipe height, h_sp (x₃).

The preliminary evaluation of the NIR grading efficiency was considered for the example of the analysis of Scots pine seeds, and the results of the preliminary experimental study are shown in

Table 3. Preliminary detection options for the choice of a plan center for Scots pine seeds’ grading using a spectrometric feature in the near-infrared-wavelength region (amplitude A₀ = 60).

Wavelength, nm	Sample Number	Seed Grading Efficiency, %
		EC	ENC
970	1	99	99
	2	100	100
	3	100	98
	4	99	100
	5	100	100
	Mean ± SD	99.6 ± 0.5	99.4 ± 0.8

. The NIR grading cycle was performed for 5 samples (n = 200 seed in each) of Scots pine seeds in two spectrometric groups: conditioned (C) seeds and non-conditioned (NC) seeds.

The results of preliminary experiments conducted on seed samples (n = 1000) showed that the efficiency of the separation of Scots pine seeds in the infrared range was 99.5 %. In this regard, the choice of engineering parameters is the result of a compromise between the cost of rejection of high-quality seeds and losses when using unrecognized low-quality seeds and is carried out according to the results of pre-experiments.

To account for the proportion of conditioned seeds that got into the receiver for non-conditioned seeds, the total grading efficiency ΔE or y-function—y (λ_r(NIR), α, h_sp)—to DOE theory is equivalent to the equation:

$$\Delta E=\frac{\Delta NC}{NC}-\frac{\Delta C_{\left(NC\right)}}{C},$$

(1)

where ΔNC is the number of eliminated non-conditioned seeds caught in the appropriate receiver (position 8 in Figure 1), NC is the number of total non-conditioned seeds (N_NC = 100), $\Delta C_{\left(NC\right)}$ is the number of conditioned seeds caught in an inappropriate receiver (position 8 in Figure 1), and C is the number of total conditioned seeds (N_C = 100).

The optimum of seed separation using a spectrometric feature in the infrared-wavelength region is equivalent to the mathematical expectation of the grading degree, tending to $Mx[y {\left({\lambda }_{r\left(\mathrm{NIR}\right)}, \alpha ,h_{sp}\right)}_{}] \to 1$ [44]. The optimum for Scots pine seeds’ grading was approximated by a second-order regression equation that takes into account linear, quadratic, and paired interactions between factors:

$$y=b_0+b_1{x}_1+b_2{x}_2+b_3{x}_3+b_{12}{x}_1{x}_2+b_{23}{x}_2{x}_3+b_{13}{x}_1{x}_3+b_{11}{x}_1^2+b_{22}{x}_2^2+b_{33}{x}_3^2$$

(2)

where b are coefficients for linear (x₁, x₂, and x₃), quadratic ($x_1^2, x_2^2,\mathrm{and} x_3^2$), and paired interactions (x₁x₂, x₂x₃, and x₁x₃) of function arguments—grader engineering factors x.

Further, based on the mathematical theory of experiment planning [43,46] and applying the procedure for conducting the experiment similarly to Novikov et al. [44], we developed and filled in the experiment matrix (

Table 4. Planning matrix for testing the efficiency of grading Scots pine seeds using a grader in the NIR wavelength region.

Experience Number	Design Matrix			Results of Experience
	Factors
	x1 (λr(NIR))	x2 (α)	x3 (hsp)	y1 (m1)	y1 (m2)	y1 (m3)	$\overline{{\mathit{y}}_1}$
1	+1	+1	+1	0.83	0.86	0.88	0.8567
2	−1	+1	−1	0.81	0.83	0.89	0.8433
3	+1	−1	−1	0.85	0.9	0.88	0.8767
4	−1	−1	+1	0.79	0.82	0.82	0.8100
5	+1	+1	−1	0.81	0.85	0.85	0.8367
6	−1	+1	+1	0.86	0.89	0.9	0.8833
7	+1	−1	+1	0.81	0.8	0.79	0.8000
8	−1	−1	−1	0.79	0.78	0.75	0.7733
9	0	0	0	1	1	0.98	0.9933
10	0	0	0	0.98	1	0.99	0.9900
11	0	0	0	1	1	0.99	0.9967
12	0	0	0	1	0.99	1	0.9967
13	0	0	0	1	1	1	1.0000
14	0	0	0	1	0.99	0.99	0.9933
15	+1.682	0	0	0.97	0.98	0.98	0.9767
16	−1.682	0	0	0.98	0.97	0.98	0.9767
17	0	+1.682	0	0.92	0.93	0.95	0.9333
18	0	−1.682	0	0.88	0.9	0.89	0.8900
19	0	0	+1.682	0.93	0.97	0.92	0.9400
20	0	0	−1.682	0.94	0.96	0.91	0.9367

). Using Microsoft Excel (Microsoft Inc.), the coefficients b of Equation (1) were determined using the least squares method and their significance was checked. Finally, we obtained an adequate (according to the Fisher criterion) regression model (see Equation (3)).

The degree of the similarity and difference between inventions and criteria for possible use (Appendix A,

Table A1. The degree of evaluation was encoded using rank variables as follows: 1—high; 2—medium; 3—low; and 4—zero (or no description, data). A Scopus request [TITLE-ABS-KEY (NIR AND “fuzzy logic” AND seed AND (sort? OR grad? OR separate?))] was used.

Inventions Title	Field	Applied Grading Algorithm	Using the Fuzzy Logic Paradigm (Crt1)	Degree Complexity of the Fuzzy Logic Algorithm (Crt2)	NIR Use (Crt3)	Complexity of the NIR Design Scheme (Crt4)	Possibility of Implementation for Grading Scots Pine Seeds (Crt5)	Grading Efficiency When Using the Algorithm (Crt6)	Invention Number
1. Devices and methods for optically identifying characteristics of material objects (Inv1)	Minimally invasive medical screening; inspection of food products; control of technological processes; study of fauna or flora; inspection of industrial materials; quality control of the production line; comparison of cosmetics and masks; environmental monitoring; identification of the liquid, powder, and solid; monitoring of the beam of liquid and gas and ore separation	Algorithms based on discriminant analysis, principal component analysis, feature extraction, Bayes theorem, genetic algorithms, decision trees, differences of least squares, artificial neural networks, wavelet theory, pattern recognition algorithms, syntactic analysis, artificial intelligence, polynomial classifiers, Walsh transformations, fuzzy logic, and fuzzy logic trees	3	4	3	4	3	4	Patent US 6,122,042 (2000)
2. Method for grouping a plurality of growth-induced seeds for commercial use or sale based on testing of each individual seed (Inv2)	Production of a separate group of growth-induced seeds, each of which has a specific characteristic, for subsequent commercial use or sale	Use of a genetic algorithm, statistical analysis, fuzzy logic, and regression methods	3	4	3	4	1	4	Patent US 8,613,158 (24 December 2013)
3. Methods of segmenting a digital image (Inv3)	Image processing and image segmentation for object-oriented classification in particular	Fuzzy logic algorithm based on linking similar pixels in the NIR region of the image	1	3	3	4	2	2	Patent US 8,233,712 (31 July 2012)
4. Spectroscopic tomography systems and methods for non-invasive detection and measurement of analytes using collision computing (Inv4)	Property or concentration of a material, changes in the amount or properties of the material, or an event or anomaly of interest and, in one example, specifically a system that performs measurement of biochemical analytes using diffuse reflectance spectroscopy	Digital filtering, stochastic filtering, auto-regression, empirical mode decomposition, Kalman filters, wavelet decomposition, matched filters, neural network, and fuzzy logic	3	4	1	1	3	4	Patent US 9,554,738 (31 January 2017)
5. Methods and systems for reducing soil compaction using worksite treatment based on determined soil properties (Inv5)	Determining soil properties	Lookup tables, fuzzy logic, neural networks, machine learning, rules-based systems	3	4	1	4	4	4	Patent Application EP 3,812,980 (28 April 2021)
6. Farm ecosystem (Inv6)	Selecting a seed with predetermined end product properties, selecting microbes to assist seed growth and providing microbes to help the seed during early-stage growth, planting the seed on a farm and periodically capturing growth data with one or more sensors, storing the growth data on a blockchain, viewing the growth data by an interested party, and controlling an irrigation system in response to the growth data	Fuzzy logic (no method specified)	2	4	2	1	2	4	Patent Application US 2,021,0271,877 (2 September 2021)
7. Smart farming (Inv7)	Agriculture irrigation systems	Images of tomato plants under a light-controlled environment analyzed using RGB color space as corresponding to N level, where a fuzzy logic control algorithm automatically adjusts the camera exposure and gain in order to control image brightness within a target gray level	3	4	1	1	3	4	Patent Application US 20,210,073,540 (11 March 2021)
8. Proposed algorithm (Inv8)	Grading of Scots pine seeds	Fuzzy logic algorithm using the Mamdani–Zadeh method	1	3	1	3	1	1	Planning future

) when creating future NIR grading methods was assessed using hierarchical cluster analysis [47], estimating the distance of criteria from the center based on the Euclidean distance square measure with IBM SPSS Statistics (IBM Corp., Armonk, NY, USA).

3. Results

Optimizing Grader Parameters

Based on the data in Table 2, a planning matrix was constructed (Table 4). The planning matrix was based on the full-factor experiment 23, six experiments in the center of the plan, and six experiments at the distance of the stellar arm, α, from the center of the experiment. Thus, we considered it quite sufficient to conduct 20 experiments to adequately describe the results of the NIR grading process.

Interpretation of the obtained NIR grading regression model

$$\begin{gathered} \mathrm{y}=0.99835+0.00439\left(\frac{{\lambda }_{r\left(\mathrm{NIR}\right)}-970}{6}\right)+0.01705\left(\frac{\alpha -45}{6}\right)- \\ -0.01583\left(\frac{{\lambda }_{r\left(\mathrm{NIR}\right)}-970}{6}\right)\left(\frac{\alpha -45}{6}\right)+0.01250\left(\frac{\alpha -45}{6}\right)\left(\frac{h_{sp}-0.2}{0.06}\right)- \\ -0.01667\left(\frac{{\lambda }_{r\left(\mathrm{NIR}\right)}-970}{6}\right)\left(\frac{h_{sp}-0.02}{0.06}\right)-0.02843{\left(\frac{{\lambda }_{r\left(\mathrm{NIR}\right)}-970}{6}\right)}_{}^2- \\ - 0.05140{\left(\frac{\alpha -45}{6}\right)}_{}^2-0.04197{\left(\frac{h_{sp}-0.2}{0.06}\right)}_{}^2 \end{gathered}$$

(3)

Made it possible to determine that changing the inclination angle from the light source significantly affects the change in the y-function due to the highest absolute values of the coefficients (0.00439) and (0.02843) in Equation (3). Further, in terms of the influence on the total grading efficiency is the mutual influence of two engineering parameters of the automatic grader—the wavelength and the height of the seed pipeline. The NIR wavelength parameter had the minimum effect on the y-function. The coefficients in Equation (3) had a negative sign, and with a decrease in these effects, the response value increases and vice versa.

The main engineering parameters of the grader were fixed in turn in the center of the plan, and based on the other two parameters (arguments), the response surface (y-function) was constructed. The response surface for fixed λ_r(NIR) is shown in Figure 2, for α in Figure 3, and for h_sp in Figure 4.

The response surface plotted in Figure 2 predicts the possibility of achieving a total grading effect in the range of 0.985–1.0 at pipeline height 0.16–0.22 m and angle 44–46°. The response surface plotted in Figure 3 predicts the possibility of achieving a total grading effect in the range of 0.985–1.0 at NIR wavelength 968–973 nm and pipeline height 0.18–0.23 m. The response surface plotted in Figure 4 predicts the possibility of achieving a total grading effect in the range of 0.985–1.0 at angle 44–46° and NIR wavelength 968–973 nm.

4. Discussion

Technological processes for improving the physiological quality (conditioning) of Scots pine seeds with the use of technical means may include grading operations for various physical and mechanical properties, rolling operations of the outer shell (coating), and encapsulation operations in combination with biophysical and biochemical activation operations. Grading, as a rule, is carried out permanently only in specialized forest seed-growing centers and nursery complexes by alternately passing through several technical means, which has an invasive effect on seeds and leads to a decrease in sowing qualities and viability. The coincidence of the density and size of dead petrified seeds with viable ones makes it impossible to eliminate them by size and density.

The technological processes of conditioning Scots pine seeds acquires the greatest relevance when they are ranked depending on the goals and objectives set by the consumer: for practical use in the implementation of technology for obtaining high-quality reproductive material in automated nurseries and forest-cultivated areas or for scientific use during breeding and genetic experiments. Moreover, the effectiveness of a group of seeding operations in the reforestation structure depends on the quality of seeds. Each forest seed under such conditions should have the maximum complex of resistance to various stressful conditions and a high viability index. The Scots pine seeds of the first quality class allow a 91–100% germination rate determined after 14 (21) days; therefore, an increase in this indicator for seeds of the first class even by 5–7% (up to 95–97%) and the separation of all viable seeds from seeds of other classes will significantly improve the seed quality and reduce seeding rates. As a result of the activities carried out, a reduction in the costs of processing, sowing, and subsequent agrotechnical measures is predicted.

The basic link uniting the technological processes of conditioning Scots pine seeds, and, as a consequence, seedlings obtained from them, is a spectrometric feature in the NIR region. An optical beam of a certain wavelength, passing or reflecting from a single seed, causes a certain response in its structure, allowing, depending on the exposure, either the detection and subsequent separation or the activation of biological properties. The spectrometric feature of the seed, both in the visible and infrared ranges, is individual and hereditarily determined and is characterized by good reproducibility of estimates (low identification error). The most active use of the spectrometric feature for the identification of forest seeds is being developed by scientists from Sweden [48,49,50], Iran [51], and China [52]. Laboratory studies were carried out on stationary spectrometers with long exposure. However, the technologies claimed to improve the quality of forest seeds still cannot enter the market. This is due to the fact that stable technological regulations for improving the quality (viability) of forest seeds based on spectrometric properties, their ranking by quality classes depending on subsequent technological operations, and recommendations for further use to obtain high-quality reproductive material are not fully developed in world seed practice. The circumstances listed here indicate the unavoidable annual shortage of high-quality reproductive material in reforestation production due to non-viable seeds being sown.

At the time of the study, we were only aware of NIR spectrometry of Scots pine seeds using stationary spectrometers. These studies, mainly by Swedish scientists, convincingly showed the possibility of separating pine seeds in the NIR range [53] and were the starting point. Moreover, joint cooperation with Swedish colleagues allowed us to establish a model for detecting a single seed of Scots pine in the NIR region [4].

Naturally, given the current level of scientific development, it is difficult to assume that NIR paradigms have never been used to separate materials, including seeds. Therefore, in early December 2021, an additional search was conducted using the Scopus platform. The request [TITLE-ABS-KEY (NIR AND “fuzzy logic” AND “Scots pine” AND seed AND (sort? OR grad? OR separate?))], which includes all the necessary keywords to identify the direction of our research, returned 0 results. We universalized the results by excluding the tree species from the query. The request [TITLE-ABS-KEY (NIR AND “fuzzy logic” AND seed AND (sort? OR grad? OR separate?))] returned 61 results for inventions. Only 8 relevant inventions were selected by a complete search and examination of a sample of 61. Many records in this sample duplicated each other (application and patent), and many did not relate to the field of agriculture and forestry due to the ambiguity of the translation of the word “seed.” We believe that it would be appropriate to provide a systematic analysis of the eight most relevant inventions (Inv1–Inv8) in Table A1 in Appendix A. For the analysis, the following rank criteria (Crt) were used, visualized on the dendrogram (Figure 5): Crt 1—using the fuzzy logic paradigm; Crt 2—degree complexity of the fuzzy logic algorithm; Crt 3—NIR use; Crt 4—complexity of the NIR design scheme; Crt 5—possibility of implementation for grading Scots pine seeds; and Crt 6—grading efficiency when using the algorithm.

The characteristic distribution of relevant inventions into three clusters (I—Inv4, Inv6, and Inv7; II—Inv1, Inv2, and Inv5; and III—Inv3 and Inv8) suggested the similarity of the proposed algorithm with the image segmentation process. The criteria for evaluating inventions were also formed by three characteristic clusters: A—Crt2 and Crt6; B—Crt1 and Crt5; and C—Crt3 and Crt4. Thus, it can be assumed that when choosing future prototypes to create an invention, the complexity of the fuzzy logic algorithm will be closely related to the grading efficiency and the use of NIR with the complexity level of the NIR design scheme.

The interpretation of the dendrogram indicates that the NIR region grading (or seg-mentation) fuzzy logic algorithms associate pixels [54,55] detected by a charge-coupled device or reflected optical beam signals detected by a photodetector [4] with similar characteristics in adjacent areas. The simplest grading method remains the pore value based on the NIR region. The threshold value (in our case, the amplitude of the reflected optical beam) is usually chosen by trial and error, for example, by using a preliminary spectrometric analysis of a large sample of seeds with known viability in advance and the following discriminant analysis [56] or by analyzing a histogram of a multispectral image [57] by a user or a given algorithm aimed at individual regions. In a more complex form, the threshold value can be set taking into account local properties (local threshold value) and the location of peaks on the spectrogram [58] or pixels in the image (adaptive threshold value) [59].

Of course, the approach to optoelectronic grading of Scots pine seeds is not only to configure the system for a specific user. First, as shown in previous studies and inventions with the participation of the authors, it consists in the novelty of the constructive optical scheme when detecting a single seed [4], increasing the energy efficiency of the device by reducing overall dimensions and increasing accuracy due to the absence of rotating optical elements [35,38,39,40] compared to photoseparators used in agriculture.

Despite the increased accuracy of the optical method compared to mechanical grading, it will not be possible to avoid fuzziness due to the inevitable deviations of the wavelength and amplitude of the optical beam used in the study.

Thus, the use of fuzzy logic allows avoiding computational difficulties associated with the use of traditional approaches based on the use of statistical methods of information processing. The use of the latter requires, first, the calculation of a significant number of statistical indicators, which increases processing time and computational costs, and second, it turns out to be incorrect in the absence of information about the nature of probability distributions of deviations of the parameters of the optical beam used.

For integrity, an ambitious goal was planned—to collect a data bank of the results of NIR grading of seeds for gymnosperm and angiosperm tree species growing in the forest. This will allow integrating the NIR grading procedure of seeds into the FLR pipeline for each reforestation initiative. This task is quite feasible, but it requires the involvement of grant funding. We are working in this direction, but we will also be glad to see in the future research by scientific groups in this direction of NIR grading of other forest tree species.

5. Conclusions

The results of this study showed:

• Automated NIR grading of Scots pine seeds is non-invasive.
• Automated NIR grading of Scots pine seeds allows the separation of seeds according to the viability index, which is important, since dead petrified seeds [48] may occur in the seed batch, which cannot be eliminated either by size or by mass.
• The Scots pine seed NIR grading efficiency in the range of 0.985–1.0 can be provided at λ_r(NIR) = 970 nm, α = 45°, and h_sp = 0.2 m of the grader’s engineering parameters.
• Cluster analysis established that the threshold value of the amplitude of the NIR-reflected optical beam on the spectrogram or the pixel characteristic on the image forms a single cluster and can be used to implement simple grading methods for Scots pine seeds.
• In the future, it is planned to develop an NIR grading system that determines the Scots pine seeds’ provenance with sufficient accuracy. It would also be quite interesting to develop and approbate this on seeds of other forest species.