1. Introduction
The last twenty years recorded a significant development of unmanned aerial vehicles (UAVs) or drones in a range of applications such as observation, geographical studies, monitoring of fires, safety, military applications, search and rescue, or agriculture [
1]. In agriculture, remote sensing by UAV appeared as very effective in the assessment of the condition of crops. Compared for example with satellite imagery, UAV monitoring features a greater accuracy of measured data [
1]. The development of small multi-rotor UAV systems is provided by their portability, low purchasing price, maneuverability, and simple use [
2].
Unmanned aerial vehicles surpass manned aerial vehicles in the versatility of use, in cost, and in the capacity of reaching finer time and space resolution [
3]. Lower flight altitude of UAV allows to use cheaper sensors compared with those that are necessary in piloted aircrafts without disturbing the upper storey of monitored crops when flying. Moreover, the speed of deployment and possibility of data collection are useful in inaccessible areas such as waterlogged lands [
4].
Yang and Hoffmann [
5] mentioned other advantages of UAV compared with satellite systems such as remote sensing of Earth platforms. These include easy deployment, availability of images for visual assessment almost in real time and overcoming of weather limitations especially as far as cloudiness is concerned. The authors also pointed out that satellite images of required target areas cannot be taken always in the necessary time due to limitations connected with satellite orbits.
The above pros predetermine the UAV for use in precision agriculture which also recorded a significant progress in recent years [
6]. There is a number of various UAV applications in this field mentioned in literature.
One of them is the monitoring of crops by means of high-resolution multispectral images. This is how various vegetation indices (VIs) can be determined, which can be used to assess the condition of plant biomass [
7], to localize weeds [
8], pests, diseases [
9,
10], to establish nutrient requirements [
7], to estimate yields [
11], or to identify drought stress [
12]. Based on the obtained information, efficient measures with the use of agrochemicals can be implemented at a given site, at a given time and in an optimum amount [
13,
14], thus saving agricultural inputs, mitigating environmental impacts, and increasing profits. UAV are selected as an option to satellite remote sensing because of their spatial and temporal resolution. However, some technical aspects can be improved to increase efficiency in precision agriculture and to evaluate economic benefits of UAV application at a farm-scale level [
2]. Moreover, law restrictions of UAV operation by governments need to be considered for their utilization in precision farming.
Matese et al. [
3] compared the normalized difference vegetation index (NDVI) obtained from three different platforms: satellite, manned aircraft, and UAV in order to evaluate possibilities of each platform in the characterization of spatial variability of experimental vineyard. The authors found out that due to their coarser spatial resolution, satellite images cannot adequately represent variability inside the vineyard with a higher heterogeneity, and based on this finding, they concluded that UAV is suitable for relatively smaller plots and that the breaking point occurs around an area of five hectares. In larger areas, other platforms appear as more advantageous.
Oil seed rape (
Brassica napus L.; OSR) is one of the most important crops in the world for human consumption, a source for meal with a high content of proteins to feed farm animals, and a renewable biological raw material for chemical and oil industry [
15]. Yield components of OSR include the number of pods, the number of seeds in the pod, and the weight of one seed [
16].
OSR passes through three different morphological forms during its development. At the first stage, the growth is dominated by leaves, at the second stage, the growth features yellow petals, and at the third stage, the growth forms stems and pods. Each form of the growth strongly affects the mode of solar radiation interception [
17]. Plants such as wheat and maize with inconspicuous flowers simply become green upon entering the reproduction growth stages, and yellow or even brown when they ripen [
11]. Many oil crops of Brassica genus, however, become green and then yellow due to the occurrence of conspicuous yellow flowers, with the green and the yellow overlapping when the growth ripens. This time-spectral variability deserves a very careful consideration when selecting a spectral index that should correlate with the physiologically significant variable such as the yield of seeds [
11]. The time-spectral variability is a function of the morphological development of OSR.
One of the most used vegetation indices is the normalized difference vegetation index (NDVI) that combines two of the most important agronomic parameters: the status of plants (occurrence of stress) and the amount of biomass on the unit area. NDVI is also used to distinguish vegetation from non-vegetation areas. Values of NDVI were found to be higher in agricultural areas than in forest areas, while water surfaces provided minus NDVI values [
18]. As it correlates well with the photosynthetic capacity, NDVI theoretically provides effective yield estimates [
19].
Another used vegetation index is the blue normalized difference vegetation index (BNDVI) which is an index without red channel availability that uses the visible blue for areas sensitive to chlorophyll content. It provides the ratio between the near infrared (NIR) and the blue zone of the spectrum, which strongly correlates with the leaf area index (LAI) [
20].
Sulik and Long [
21] suggested a normalized difference yellowness index (NDYI) determining a normalized ratio between the Green and the Blue zone. It was proposed in order to estimate the number of flowers in OSR and other yellow-flowering plants and hence to derive the vegetation cover density. Production of OSR flowers plays an important role in yield prediction, and the yellowness of OSR petals can be a critical reflective signal and a good predictor of the number of pods, and hence the yield of seeds [
16]. It is a known fact that yellow flowers increase reflectance of green color while slightly decreasing reflectance of blue color [
22]. Therefore, the blue zone is a good denominator for the index of flower variability, the value of which increases with the increasing reflectance in the green zone. For the numerator, the green zone is a better choice than the red zone because the red color reflectance changes with the content of chlorophyll which changes with the displacement of soluble photosynthate by healthy leaves into buds and flowers [
23,
24]. This is why the ratio of blue and green light should be functioning properly for the estimate of yellowness as it changes in opposite directions due to the presence of yellow flowers. The green/blue ratio values should have a positive relation to the density of flowers while the NDVI values decrease with the increasing coverage of flowers [
23].
It was also found [
24] that NDYI is the most useful index of all studied VI for the monitoring of changes in the content of chlorophyll in leaves as well as the greenness of rice leaves during the entire period of growth.
The relationship between some VI and OSR yields was studied already by many authors [
16,
25,
26,
27]. New benefits of this research are proposals of prediction models based on which future OSR yields can be estimated with high accuracy at the stage of flowering even in the conditions of high variability of above-ground and under-ground vegetation environments caused by various biotic and abiotic stresses.
The goal of this study was to verify, on the basis of regression models, the suitability of NDVI, BNDVI, and NDYI vegetation indices obtained by means of UAV at the stage of flowering for predicting the OSR yield, and the usability of the vegetation indices for the identification of anomalies in the condition of the flowering OSR growth, these being the initial phases of a long-term field experiment focused on vegetation indices potentially applicable in studying winter OSR at the stage of flowering.