1. Introduction
Increasing the acreage under soybeans is a solution to the problem of the shortage of high-grade vegetable protein in the Russian Federation. Soybeans are actively spreading in the northwest direction, and until now the main centers of their cultivation were the Far Eastern, North Caucasian, and Central Federal Districts [
1]. Changing climatic conditions also require the search and creation of adapted varieties [
2]. Therefore, it is necessary to study the plants of such varieties during the phenology stages, namely, stage I (seed germination) and stage II (germination), which are especially important since the further development of the plant depends on them [
3].
To fully realize the genetic potential of cultivated varieties of agricultural crops, it is important to ensure the high quality of their seeds [
4]. For this purpose, in accordance with the Federal Law “On Seed Production” [
5], seeds intended for sowing were subjected to inspection for varietal and sowing qualities during varietal and seed control. High-quality seeds have high indicators of germination energy, laboratory germination, and growth strength, which ensures the production of friendly seedlings and high field germination. If the seeds have low quality indicators, there is an increase in sparse crops and the formation of plants with low productivity.
There are many factors affecting the germination energy and field germination of seeds, such as soil temperature, soil granulometric composition, the depth and uniformity of seed embedding in the soil, the presence of moisture, the appearance of disease, and others. The soybean is a thermophilic culture by its biology; the minimum temperature for seed germination is +8 to +10 °C, with +10 °C being more favorable, and the temperature range of +12 to +14 °C contributes to the appearance of friendly seedlings [
6,
7]. The increase in field germination leads to a properly selected depth for sowing seeds and their uniform embedding in the soil [
8]. A large depth of seed deepening leads to an elongation of the germination period and there is a risk of soil infection. Conversely, insufficient deepening leads to uneven seedlings and the production of weakened plants. An essential condition for the emergence of seedlings is the presence and amount of moisture in the soil. Seeds require at least 130–160% of their weight in water to germinate [
7,
9]. Weather conditions directly affect the amount of moisture in the soil. Prolonged rains or periods of drought adversely affect germination. Thus, field germination is a complex indicator that includes the quality of seeds, the level of agricultural technology, soil composition, weather conditions, etc. Due to the variety of factors affecting the field germination of soybean plants, researchers did not identify a relationship between laboratory germination and field germination [
10,
11].
Field germination is estimated by counting the germinated plants per unit area and expressing it as a percentage of the sown seeds. However, this method requires not only certain physical costs but also time. With the development of modern digital technologies, it has become possible to search for new methods to assess field germination [
12,
13].
Remote sensing methods are used to map vegetation in the agricultural sector [
14]. In recent years, unmanned aerial vehicles (UAVs) have attracted considerable interest in solving problems in the breeding process [
15,
16,
17]. In particular, UAVs with optical suspension equipment have become useful tools in the evaluation of breeding plots [
18,
19,
20]. The use of UAVs makes it possible to study breeding crops and obtain objective data on several signs and physiological qualities of studied crops [
21,
22,
23]. In the study of soybean plants, various vegetation indices are used [
24,
25,
26] for phenotyping [
27], yield forecasting [
28,
29,
30], flood stress assessment [
31], drought [
32], detection of weeds [
33], and fertilizing [
34]. The issue of assessing the field germination of soybean crops using multispectral data was not sufficiently disclosed, and its further study is relevant.
In connection with the above, the purpose of our research was as follows: to develop field germination ranges for breeding soybean plants according to multispectral survey data from an unmanned aerial vehicle.
3. Results
On 11 June 2020, 13 June 2021, and 14 June 2022, monitoring was carried out with a UAV and a multispectral camera. Flight parameters for each year are presented in
Table 1. The altitude, speed of flight, and overlap of images did not change throughout the study.
RGB and MSD images were collected during each flight. During the three years of the study, 2696 images were collected. The volume of raw data was 14,163 Mb. The total data volume of the current study was 32,253 Mb [
48]. The total amount of data considered the volume of raw data, the size of PIX4Dmapper software projects and the created digital maps (
Table 2).
After the flight, an orthophotoplane was created, and the vegetation indices NDVI, NDRE, and ClGreen were calculated (
Figure 6). The accuracy of the projects was evaluated for RGB and MSD. There were six projects. The RMSEp calculation was performed after the initial aerotriangulation and marking of all GCPs. The project error did not exceed three centimeters for all projects according to the PIX4Dmapper Report.
After calculating the average value of vegetation indices and the percentage of germination for each of the plots, a correlation analysis of ground and air data was carried out. Correlation analysis of the obtained germination data and values of vegetation indices (
Figure 7):
for germination and NDVI, r = 0.72;
for germination and NDRE, r = 0.70;
for germination and ClGreen, r = 0.75.
The obtained indicators show the presence of a positive relationship of an average degree between the data.
As a result, 1460 plots were divided into two groups: test and verification. The test group contained 744 plots and the verification group contained 716 plots. For the test plots, the number of germination groups according to the Sturges’ rule was (10):
The interval step was (11):
Ranges for eleven groups of field germination were obtained (
Table 3), where the first group was the lowest germination, and the eleventh group was the highest germination. For low germination, the minimum value is assumed to be 0%, and the maximum value for high germination is 100%.
There is a frequency of distribution of field germination in the eleven groups shown in
Figure 8. The main distribution falls in groups 6–8. When assessing the degree of asymmetry, an insignificant (
p = 0.05) left-sided asymmetry was revealed. In this case, the deviation from the normal distribution was considered insignificant, with SEx (mean square error of the kurtosis coefficient) < 3 (SEx = 0.768).
To simplify the assessment of field germination, a grouping was created with the following ranges:
Group 1 (1): very low germination: 0–12.84;
Group 2 (2–3): low germination: 12.85–31.14;
Group 3 (4–6): average germination: 31.15–58.59;
Group 4 (7–11): high germination: 58.6–100.
Of the 744 plots, 9 plots had very low germination; 30 plots had low germination; 245 plots had average germination; and 460 plots had high germination. In accordance with
Section 2.4, the Sturges’ rule was applied to identify ranges of vegetation indices (
Table 4).
We assigned a value for the data of each of the groups for ground and air research: very low germination: 0; low germination: 1; average germination: 2; and high germination: 3. Indicators for 84 validation plots for 2020, 184 plots for 2021, and 448 plots for 2022 were also assigned values. The MAPE error showed the accuracy of the obtained results (
Table 5). The final value of MAPE was calculated under the condition that the average values of vegetation indices fell into at least two of the three ranges, and then the appropriate germination level was assigned. If this condition was not met, then the germination level was assigned below the one under consideration.
The percentage of error did not exceed 10% for the vegetation indices NDVI and ClGreen, which indicates a high accuracy of the data obtained. For the NDRE vegetation index, the error percentage was higher and reached 18%. The smallest error of 2.98% was detected in 2022 for ranges of NDVI values. The final values of MAPE for three years did not exceed 10%.
Germination maps were created for the 2020–2022 years using software for automatic evaluation of the germination of soybean crops (
Figure 9). Based on the results of the program, the number of plots for each of the germination levels for three years for verification plots was calculated (
Table 6). Evaluation of the accuracy of the software was carried out relative to the data obtained by breeders. The maximum error of the software did not exceed 10% for each of the germination levels.
4. Discussion
As a result of the three-year study, vegetation maps of three vegetation indices (NDVI, NDRE, and ClGreen) of soybean breeding crops in the full germination phase were created. The project error along the X, Y, and Z axes did not exceed 3 cm (RMSE
p < 3 cm). This indicated a high point in the construction of digital maps (orthophotoplane and vegetation maps) [
49,
50].
Comparative analysis of field germination using the method of vegetation indices and by conducting ground measurements showed their general patterns. Thus, the use of the values of the vegetation indices NDVI, NDRE, and ClGreen revealed a correlation with the ground assessment of field germination (
Table 2). The correlation coefficients here were quite close and amounted to r = 0.70–0.75, which may indicate the applicability in the calculations of each of the presented vegetation indices. The average absolute percentage error has been calculated in many studies on the use of UAVs in agriculture [
51,
52,
53,
54,
55,
56,
57,
58,
59]. Studies with a high level of accuracy consider MAPE values up to 15% [
52,
56,
57,
58,
59]. In the current study, the MAPE values did not exceed 10% for two vegetation indices (NDVI, ClGreen) and was 18% for one vegetation index (NDRE). The overall MAPE for every year did not exceed 10%, which indicated high accuracy of the results of the germination assessment using multispectral data from the UAV. The presence of an error of approximately 10% was possible due to many factors, for example, the use of a breeding seeder without an accurate seeding system [
60], varietal characteristics, soil, and climatic conditions.
The developed software allowed us to evaluate the germination of soybean crops according to three vegetation indices. The greatest errors were observed for the germination levels of average and high germination. Under favorable ecological and climatic conditions, the largest number of soybean plots had an average or high germination rate. The difficulty in assessing germination lies in finding the correct ranges for each of the germination levels. Despite all the difficulties, the software showed high accuracy, providing an error rate of less than 10% for each of the groups.
To be able to repeat the experiment, aerial and ground surveys were carried out every year only during the full germination phase. Aerial photography was carried out according to the recommendations for collecting multispectral data (
https://support.micasense.com/hc/en-us/articles/224893167-Best-Practices-Collecting-Data-with-MicaSense-Sensors, accessed on 25 April 2020). The calibration panel and DLS 2 for a MicaSense Altum were used. We did not set the task to compare the numerical indicators of field germination with the values of vegetation indices; therefore, the study used the percentage values of field germination relative to the seeding rate. The germination rate was influenced by weather conditions at the time of seed germination, including air temperature and precipitation, regardless of the location of sowing.
The size of the breeding plots was dictated by the requirements for the methodology of field experience. This technique is standardized in the Russian Federation [
35,
36,
37]. The experience shown in this article is an additional study of the material used for breeding work. The statistical power of this study was achieved by combining small plots into larger groups. At this phase of plant development, the marginal effect was insignificant, and the entire selection of breeding plots reflected the properties of the general population of soybean plants regardless of variety, such as the place of sowing or other environmental factors.
The advantages of the survey of breeding crops with the help of UAVs were high efficiency and productivity; however, the reliability of the information received and the ability to assess it in the field was difficult. The method proposed in this study for assessing the field germination of breeding plots was considered an additional tool for the breeder to assess the development of new varieties. In the future study of additional vegetation indices, special attention will be paid to the vegetation indices calculated on the blue channel. Additional research is needed to develop a new methodology as the breeder’s main tool for assessing the field germination of soybean breeding seedlings.
5. Conclusions
As a result of this study, the field germination data of soybean plants for three years (1460 plots) and the average values of the vegetation indices NDVI, NDRE, and ClGreen were analyzed. The correlation coefficients of vegetation indices and ground-based assessment (r = 0.70–0.75) indicated an average positive relationship between multispectral and ground-based studies. The ranges of field germination (very low, low, average, and high field germination) were calculated for the ground data and the vegetation indices NDVI, NDRE, and ClGreen. The values of the average absolute percentage error for the test plots in 2020–2022 did not exceed 10% for the NDVI and ClGreen indices, or 18% for the NDRE, and the overall MAPE was less than 10%, indicating a high level of accuracy of the obtained germination ranges. The ratios of germination ranges and values of vegetation indices were revealed, forming the basis of the developed software for assessing soybean germination. The percentage of software errors for each germination level did not exceed 10%. This software will serve as a tool for assessing the field germination of soybean breeding plots in 2023–2024 in the fields of the FGBNU FNC ZBK.
Thus, the evaluation of field germination of soybean breeding crops using the vegetation indices NDVI, NDRE, and ClGreen can be an independent method. The use of this method makes it possible to evaluate the field germination of a large number of experimental plots and the influence of genotype and various elements of agrotechnology on this indicator.